AVhndZddlZddlmZddlmZddlmZddlm Z ddl m Z ddlmZdd l mZdd l mZdd l mZdd lmZdd lmZddlmZddlmZm Z m!Z!ddl"m#Z#edddddddddddddddd d!d"d#d$Z$dd%e#e!ejJfd&e#e!ejLfd'e#e!e$ffd(Z'ed)ejPe'Z)d%e#e!ejJfd&e#e!ejLfd'e#e!e$ffd*Z*dd%e#e!ejJfd+e#e!ejVffd,Z,ed-ejPe,Z-d%e#e!ejJfd+e#e!ejVffd.Z.dd%e#e!ejJfd/e#e!ejLffd0Z/ed1ejPe/Z0d%e#e!ejJfd/e#e!ejLffd2Z1ed3ddddddddddddddd d!d"d#d$Z2dd%e#e!ejJfd4e#e!ejVfd5e2d+e#e!e2ffd6Z3ed7ejPe3Z4d%e#e!ejJfd4e#e!ejVfd5e2d+e#e!e2ffd8Z5gd9d:d:dfd;e6dZ8ed?ejPe8Z9d;e6ddd%e#e!ejJfd+e#e!ejVffdEZ?edFejPe?Z@d%e#e!ejJfd+e#e!ejVffdGZAedHZBdd%e#e!ejJfdIe#e!ejJfdJe#e!eBfdKe6fdLZCedMejPeCZDd%e#e!ejJfdIe#e!ejJfdJe#e!eBfdKe6fdNZEdd%e#e!ejJfd+e#e!ejVffdOZFedPejPeFZGd%e#e!ejJfd+e#e!ejVffdQZHejdRgdSZJdd%e#e!ejJfdTe#e!ejVfdUe;dVe;dWe6f dXZKedYejPeKZLd%e#e!ejJfdTe#e!ejVfdUe;dVe;dWe6f dZZMed[ddddddddddddddd d!d"d#d$ZNdd5eNdZdd%e#e!ejfd/ed+e#e!effd?Zed@ejPeZd%e#e!ejfd/ed+e#e!effdAZ edBZ!dd%e#e!ejJfdCe#e!e!fdDe;d+e#e!e!ffdEZ"edFejPe"Z#d%e#e!ejJfdCe#e!e!fdDe;d+e#e!e!ffdGZ$edHZ%dd%e#e!ejfdCe#e!e%fdDe;d+e#e!e%ffdIZ&edJejPe&Z'd%e#e!ejfdCe#e!e%fdDe;d+e#e!e%ffdKZ(edLZ)ddMe#e!ejVfd/e)d0e7d+e#e!ejffdNZ*edOejPe*Z+dMe#e!ejVfd/e)d0e7d+e#e!ejffdPZ,dd;e6de6dd%e#e!ejJffdhZ?dd%e#e!ejJffdiZ@edjejPe@ZAd%e#e!ejJffdkZBdd%e#e!ejffdlZCedmejPeCZDd%e#e!ejffdnZEejdod|dpgZFedqZGdd%e#e!ejJfdre#e!ejfd5eGfdsZIedtejPeIZJd%e#e!ejJfdre#e!ejfd5eGfduZKejdvd|dpgZLedwZMdd%e#e!ejJfdre#e!ejfd5eMfdxZNedyejPeNZOd%e#e!ejJfdre#e!ejfd5eMfdzZPejd{d|dpgZQed|ZRdd%e#e!ejfdre#e!ejfd5eRfd}ZSed~ejPeSZTd%e#e!ejfdre#e!ejfd5eRfdZUedZVdd%e#e!ejJfdme#e!ejVfdre#e!ejfd5eVd+e#e!eVff dZWedejPeWZXd%e#e!ejJfdme#e!ejVfdre#e!ejfd5eVd+e#e!eVff dZYedZZdd%e#e!ejJfdme#e!ejVfdre#e!ejfd5eZd+e#e!eZff dZ[edejPe[Z\d%e#e!ejJfdme#e!ejVfdre#e!ejfd5eZd+e#e!eZff dZ]edZ^dd%e#e!ejfdme#e!ejVfdre#e!ejfd5e^d+e#e!e^ff dZ_edejPe_Z`d%e#e!ejfdme#e!ejVfdre#e!ejfd5e^d+e#e!e^ff dZadd%e#e!ejJfdre#e!ejfde7d+e#e!ejJffdZbedejPebZcd%e#e!ejJfdre#e!ejfde7d+e#e!ejJffdZddd%e#e!ejJfdre#e!ejfde7d+e#e!ejJffdZeedejPeeZfd%e#e!ejJfdre#e!ejfde7d+e#e!ejJffdZgejdddgZhdd%e#e!ejfdre#e!ejfde7fdZiedejPeiZjd%e#e!ejfdre#e!ejfde7fdZkejdddgZldd%e#e!ejfdre#e!ejfde#e!ejVfde7fdZmedejPemZnd%e#e!ejfdre#e!ejfde#e!ejVfde7fdZoedZpdd%e#e!ejJfdre#e!ejfd5epd+e#e!epffdZqedejPeqZrd%e#e!ejJfdre#e!ejfd5epd+e#e!epffdZsedZtdd%e#e!ejJfdWe#e!ejVfdre#e!ejfd5etd+e#e!etff dZuedejPeuZvd%e#e!ejJfdWe#e!ejVfdre#e!ejfd5etd+e#e!etff dZwedZxdd%e#e!ejJfdWe#e!ejVfdre#e!ejfd5exd+e#e!exff dZyedejPeyZzd%e#e!ejJfdWe#e!ejVfdre#e!ejfd5exd+e#e!exff dZ{edZ|dd%e#e!ejfdWe#e!ejVfdre#e!ejfd5e|d+e#e!e|ff dZ}edejPe}Z~d%e#e!ejfdWe#e!ejVfdre#e!ejfd5e|d+e#e!e|ff dZedZdd%e#e!ejJfdme#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedejPeZd%e#e!ejJfdme#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedZdd%e#e!ejJfdme#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedejPeZd%e#e!ejJfdme#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedZdd%e#e!ejfdme#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedejPeZd%e#e!ejfdme#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZdd%e#e!ejJfdre#e!ejfd+e#e!ejVffdZedejPeZd%e#e!ejJfdre#e!ejfd+e#e!ejVffdZdd%e#e!ejJfdre#e!ejfd+e#e!ejVffdZedejPeZd%e#e!ejJfdre#e!ejfd+e#e!ejVffdZdd%e#e!ejfdre#e!ejfd+e#e!ejVffdZedejPeZd%e#e!ejfdre#e!ejfd+e#e!ejVffd„ZedÎZdd%e#e!ejJfd|e#e!efdpe#e!ejLfdre#e!ejfd+e#e!ejff dĄZedūejPeZd%e#e!ejJfd|e#e!efdpe#e!ejLfdre#e!ejfd+e#e!ejff dƄZedǎZdd%e#e!ejJfd|e#e!efdpe#e!ejLfdre#e!ejfd+e#e!ejff dȄZedɫejPeZd%e#e!ejJfd|e#e!efdpe#e!ejLfdre#e!ejfd+e#e!ejff dʄZedˎZdd%e#e!ejfd|e#e!efdpe#e!ejLfdre#e!ejfd+e#e!ejff d̄ZedͫejPeZd%e#e!ejfd|e#e!efdpe#e!ejLfdre#e!ejfd+e#e!ejff d΄ZedώZdd%e#e!ejJfd|e#e!efdre#e!ejfd+e#e!ejffdЄZedѫejPeZd%e#e!ejJfd|e#e!efdre#e!ejfd+e#e!ejffd҄ZedӎZdd_e#e!ejVfd5ed`e;dae;dbe7d+e#e!ejJff dԄZedիejPeZd_e#e!ejVfd5ed`e;dae;dbe7d+e#e!ejJff dքZejdd%dgZedَZdd_e#e!ejVfd5ed`e;dae;de;dbe7f dۄZedܫejPeZd_e#e!ejVfd5ed`e;dae;de;dbe7f d݄ZedގZdd%e#e!ejJfdWe#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff d߄ZedejPeZd%e#e!ejJfdWe#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedZdd%e#e!ejJfdWe#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedejPeZd%e#e!ejJfdWe#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedZdd%e#e!ejfdWe#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZedejPeZd%e#e!ejfdWe#e!ejVfd|e#e!efdre#e!ejfd+e#e!ejff dZdd;e6de6d= 1`. The type of each component in a value. shapes: An optional list of shapes (each a `tf.TensorShape` or list of `ints`). Defaults to `[]`. The shape of each component in a value. Each shape must be 1 in the first dimension. The length of this attr must be the same as the length of component_types. capacity: An optional `int`. Defaults to `-1`. The capacity of the barrier. The default capacity is MAX_INT32, which is the largest capacity of the underlying queue. container: An optional `string`. Defaults to `""`. If non-empty, this barrier is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this barrier will be shared under the given name across multiple sessions. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Cbarrier op does not support eager execution. Arg 'handle' is a ref.zEExpected list for 'component_types' argument to 'barrier' Op, not %r.component_typesNzY/Y ^ F FT5M *  "( ) **9? ?"H  H - ?& ? H   x 4(I ;7)K!!+}=+'88?6$ *7!QX QK' ""$.?!@(ll8$j +[ll;'ll=) +F ::L <2 (' .AZ @ 2G%=G*zraw_ops.Barrierctd)Nrsr?rtrurorprqr.r@s r;barrier_eager_fallbackrsZ[[r=cancel_pending_enqueuesctjxstj}|j}|jr t d|d}t j |d}tjd|||\}}}}|S)aCloses the given barrier. This operation signals that no more new elements will be inserted in the given barrier. Subsequent InsertMany that try to introduce a new key will fail. Subsequent InsertMany operations that just add missing components to already existing elements will continue to succeed. Subsequent TakeMany operations will continue to succeed if sufficient completed elements remain in the barrier. Subsequent TakeMany operations that would block will fail immediately. Args: handle: A `Tensor` of type mutable `string`. The handle to a barrier. cancel_pending_enqueues: An optional `bool`. Defaults to `False`. If true, all pending enqueue requests that are blocked on the barrier's queue will be canceled. InsertMany will fail, even if no new key is introduced. name: A name for the operation (optional). Returns: The created Operation. Ibarrier_close op does not support eager execution. Arg 'handle' is a ref.Fr BarrierCloser'rr. r0rr1r2r3rI make_boolr4r5r'rr.r6r7r8r9r:s r; barrier_closers*    0h..0$ #\\ b cc$#$../FHab'88v0G!#!QX *r=zraw_ops.BarrierClosectd)Nrr?r'rr.r@s r;barrier_close_eager_fallbackr`aar=cXtjxstj}|j}|jr t dt j d||\}}}}|dd}tjr&d}|j} tjd| |||\}|S)zComputes the number of incomplete elements in the given barrier. Args: handle: A `Tensor` of type mutable `string`. The handle to a barrier. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. Sbarrier_incomplete_size op does not support eager execution. Arg 'handle' is a ref.BarrierIncompleteSizerFNrGrHrMs r;barrier_incomplete_sizer"s    0h..0$ #\\ l mm'88T;!QX QK' ""$ F::L vw@ (' .r=zraw_ops.BarrierIncompleteSizectd)Nrr?rSs r;&barrier_incomplete_size_eager_fallbackr?jkkr=) TV_BarrierInsertMany_Tr _atypes.Boolrr_atypes.Float16rr_atypes.Float8e4m3b11fnuz_atypes.Float8e4m3fn_atypes.Float8e4m3fnuz_atypes.Float8e5m2_atypes.Float8e5m2fnuzrrr _atypes.Int4rrrrr r!r"_atypes.Resource_atypes.Stringr#r$ _atypes.UInt4r%r&_atypes.Variantkeysvaluescomponent_indexctjxstj}|j}|jr t dt j |d}tjd|||||\}}}} |S)alFor each key, assigns the respective value to the specified component. If a key is not found in the barrier, this operation will create a new incomplete element. If a key is found in the barrier, and the element already has a value at component_index, this operation will fail with INVALID_ARGUMENT, and leave the barrier in an undefined state. Args: handle: A `Tensor` of type mutable `string`. The handle to a barrier. keys: A `Tensor` of type `string`. A one-dimensional tensor of keys, with length n. values: A `Tensor`. An any-dimensional tensor of values, which are associated with the respective keys. The 0th dimension must have length n. component_index: An `int`. The component of the barrier elements that is being assigned. name: A name for the operation (optional). Returns: The created Operation. Obarrier_insert_many op does not support eager execution. Arg 'handle' is a ref.rBarrierInsertMany)r'rrrr. r0rr1r2r3rIr~r4r5) r'rrrr.r6r7r8r9r:s r;barrier_insert_manyrDs},    0h..0$ #\\ h ii%%o7HI/'88Ff-<4I!QX *r=zraw_ops.BarrierInsertManyctd)Nrr?)r'rrrr.r@s r;"barrier_insert_many_eager_fallbackrgfggr=cXtjxstj}|j}|jr t dt j d||\}}}}|dd}tjr&d}|j} tjd| |||\}|S)zComputes the number of complete elements in the given barrier. Args: handle: A `Tensor` of type mutable `string`. The handle to a barrier. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. Nbarrier_ready_size op does not support eager execution. Arg 'handle' is a ref.BarrierReadySizerFNrGrHrMs r;barrier_ready_sizerjs    0h..0$ #\\ g hh'8866!QX QK' ""$ F::L L&'; (' .r=zraw_ops.BarrierReadySizectd)Nrr?rSs r;!barrier_ready_size_eager_fallbackreffr=BarrierTakeMany)indicesrr num_elementsallow_small_batchwait_for_incomplete timeout_msc Xtjxstj}|j}|jr t dt |t tfstd|z|D cgc]} tj| d}} |d}tj|d}|d}tj|d}|d}tj|d }tjd ||||||| \} } } } | dd} tjrjd| j!dd| j#dd| j#dd | j%d f}| j&}tj(d ||| | dd | d dgz} t*j-| } | Scc} w) aTakes the given number of completed elements from a barrier. This operation concatenates completed-element component tensors along the 0th dimension to make a single component tensor. Elements come out of the barrier when they are complete, and in the order in which they were placed into the barrier. The indices output provides information about the batch in which each element was originally inserted into the barrier. Args: handle: A `Tensor` of type mutable `string`. The handle to a barrier. num_elements: A `Tensor` of type `int32`. A single-element tensor containing the number of elements to take. component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a value. allow_small_batch: An optional `bool`. Defaults to `False`. Allow to return less than num_elements items if barrier is already closed. wait_for_incomplete: An optional `bool`. Defaults to `False`. timeout_ms: An optional `int`. Defaults to `-1`. If the queue is empty, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (indices, keys, values). indices: A `Tensor` of type `int64`. keys: A `Tensor` of type `string`. values: A list of `Tensor` objects of type `component_types`. Mbarrier_take_many op does not support eager execution. Arg 'handle' is a ref.zOExpected list for 'component_types' argument to 'barrier_take_many' Op, not %r.rtNFrrrmrr)r'rrtrrrr.)r0rr1r2r3ryrzr{r|rIrfrr~r4r5rJr_get_attr_boolrrKrL_BarrierTakeManyOutput_make)r'rrtrrrr.r6r7rr8r9r:rNrOrPs r;barrier_take_manyrsF    0h..0$ #\\ f gg OdE] 3  *,; < ==JYY2X'',=>Y/Y(():@UVJ  \:*'88&|+:->/B&0t =!QX QK' ""$.?!@!3#5#56I#J#  !67 - /F ::L <: BQK712;- '' " ( ( 1' .9Zs2F'zraw_ops.BarrierTakeManyctd)Nrr?)r'rrtrrrr.r@s r; barrier_take_many_eager_fallbackrdeer=TV_ConditionalAccumulator_dtypereduction_typec tjxstj}|j}|jr t dt j |d}t j|d}|d}t j|d}|d}t j|d}|d}t j|d }tjd |||||| \}}} } | dd} t jr{d| jdd| jdd| jdd| jdd | jd f } | j} t jd | | | | \} | S) aZA conditional accumulator for aggregating gradients. The accumulator accepts gradients marked with local_step greater or equal to the most recent global_step known to the accumulator. The average can be extracted from the accumulator, provided sufficient gradients have been accumulated. Extracting the average automatically resets the aggregate to 0, and increments the global_step recorded by the accumulator. Args: dtype: A `tf.DType` from: `tf.float32, tf.float64, tf.int32, tf.uint8, tf.int16, tf.int8, tf.complex64, tf.int64, tf.qint8, tf.quint8, tf.qint32, tf.bfloat16, tf.qint16, tf.quint16, tf.uint16, tf.complex128, tf.half, tf.uint32, tf.uint64`. The type of the value being accumulated. shape: A `tf.TensorShape` or list of `ints`. The shape of the values, can be [], in which case shape is unknown. container: An optional `string`. Defaults to `""`. If non-empty, this accumulator is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this accumulator will be shared under the given name across multiple sessions. reduction_type: An optional `string` from: `"MEAN", "SUM"`. Defaults to `"MEAN"`. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Sconditional_accumulator op does not support eager execution. Arg 'handle' is a ref.r`shapeNrnrprqMEANrConditionalAccumulatorr`rrprqrr.r0rr1r2r3rIrfr}rr4r5rJrgrrKrLr`rrprqrr.r6r7r8r9r:rNrOrPs r;conditional_accumulatorrsy6    0h..0$ #\\ l mm   UG ,%   eW -%I ;7)K!!+}=+N$$^5EF.'88 U,5.91?d L!QX QK' ""$s))'2Gll7#[#,,{2KS\\-8:Jll+,.F::L  ,A (' .r=zraw_ops.ConditionalAccumulatorctd)Nrr?r`rrprqrr.r@s r;&conditional_accumulator_eager_fallbackrrr=ctjxstj}|j}|jr t j |d||}|Stjd||\}}}}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||S#t j$rYwxYw)zDelete the tensor specified by its handle in the session. Args: handle: A `Tensor` of type `string`. The handle for a tensor stored in the session state. name: A name for the operation (optional). Returns: The created Operation. DeleteSessionTensorNr.r@rF)r0rr1r2rTFE_Py_FastPathExecute_core_NotOkStatusException_opsraise_from_not_ok_status_FallbackException$delete_session_tensor_eager_fallback_SymbolicExceptionr4r5 r'r.r6r7rNer8r9r:s r;delete_session_tensorr"s    0h..0$ #\\ 11 #T63g n(88f49!QX *  & &- ##At,,  # #    1 t''  # #   0A99C B''C?C CC('C(zraw_ops.DeleteSessionTensorctj|tj}|g}d}t j dd||||}d}|S)NsDeleteSessionTensorrrKattrsr@r.rconvert_to_tensor_dtypesstringrIrr'r.r@rPrOrNs r;rrDsL  ! !&'.. 9&, &   3Q|#)s ?' ' .r=) TV_DynamicPartition_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rdynamic_partitiondata partitionsnum_partitionsc tjxstj}|j}|jr t j |d|||d|}|St||||fd}|tur|St/j0|d} t3j4d||||\}}} } | dd}t/j6rHd| j9dd| j;df} | j<} t/j>d| | ||S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||fd}|tur|St|||||S#t j$rY$tt f$rJt#j$t&dt)||||}|t"j*j,ur|cYSwxYw#tt f$rJt#j$t&dt)||||}|t"j*j,ur|cYSwxYw)a2Partitions `data` into `num_partitions` tensors using indices from `partitions`. For each index tuple `js` of size `partitions.ndim`, the slice `data[js, ...]` becomes part of `outputs[partitions[js]]`. The slices with `partitions[js] = i` are placed in `outputs[i]` in lexicographic order of `js`, and the first dimension of `outputs[i]` is the number of entries in `partitions` equal to `i`. In detail, ```python outputs[i].shape = [sum(partitions == i)] + data.shape[partitions.ndim:] outputs[i] = pack([data[js, ...] for js if partitions[js] == i]) ``` `data.shape` must start with `partitions.shape`. For example: ```python # Scalar partitions. partitions = 1 num_partitions = 2 data = [10, 20] outputs[0] = [] # Empty with shape [0, 2] outputs[1] = [[10, 20]] # Vector partitions. partitions = [0, 0, 1, 1, 0] num_partitions = 2 data = [10, 20, 30, 40, 50] outputs[0] = [10, 20, 50] outputs[1] = [30, 40] ``` See `dynamic_stitch` for an example on how to merge partitions back.
Raises: * `InvalidArgumentError` in following cases: - If partitions is not in range `[0, num_partiions)` - If `partitions.shape` does not match prefix of `data.shape` argument. Args: data: A `Tensor`. partitions: A `Tensor` of type `int32`. Any shape. Indices in the range `[0, num_partitions)`. num_partitions: An `int` that is `>= 1`. The number of partitions to output. name: A name for the operation (optional). Returns: A list of `num_partitions` `Tensor` objects with the same type as `data`. DynamicPartitionrN)rr.r@rG)rrrr.T) r0rr1r2rrrrrrr!_dispatcher_for_dynamic_partitionNotImplemented dynamic_partition_eager_fallbackrr| ValueError _dispatchr rdict OpDispatcher NOT_SUPPORTEDrIr~r4r5rJrrgrKrL) rrrr.r6r7rNrr8r9r:rOrPs r;rrPskz    0h..0$ #\\ 11  $j:Jgn20 z>4149Gn$ n$$^5EF.  )::*+9FAq#x QK' ""$ 1 12B CS  %'F::L L&'; .e  & &- ##At,,  # #    1 ^T 3T;g  & -  >  # #  z " "" r4Tj7E-1$3 g  ..<< <  ( Z    R4J5C+/"1 G i,,::: n  sPDG6E#D>>EEF5FG3AG31G36AI Izraw_ops.DynamicPartitioncVtj|d}tj|g|g\}\}tj|t j }||g}d|d|f}tjd|||||}tjrtjd||||S)NrrsDynamicPartitionrr) rIr~args_to_matching_eagerrrrint32rrJrL) rrrr.r@_attr_TrPrOrNs r;rrs$$^5EF.44dVS"E'7D%%j'--@* #, nc7 ;&   0.$0C"& ('""$ L&'; .r=) TV_DynamicStitch_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rdynamic_stitchrc dtjxstj}|j}|jr t j |d|||}|St|||fd}|tur|St/|t0t2fstd|zt5|}t/|t0t2fstd|zt5||k7rt!dt5||fz t7j8d|||\}}} } | dd}t;j<rHd | j?d d | jAd f} | jB} t;jDd| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||fd}|tur|St||||S#t j$rYtt f$rIt#j$t&dt)|||}|t"j*j,ur|cYSwxYw#tt f$rIt#j$t&dt)|||}|t"j*j,ur|cYSwxYw) a+ Interleave the values from the `data` tensors into a single tensor. Builds a merged tensor such that ```python merged[indices[m][i, ..., j], ...] = data[m][i, ..., j, ...] ``` For example, if each `indices[m]` is scalar or vector, we have ```python # Scalar indices: merged[indices[m], ...] = data[m][...] # Vector indices: merged[indices[m][i], ...] = data[m][i, ...] ``` Each `data[i].shape` must start with the corresponding `indices[i].shape`, and the rest of `data[i].shape` must be constant w.r.t. `i`. That is, we must have `data[i].shape = indices[i].shape + constant`. In terms of this `constant`, the output shape is merged.shape = [max(indices) + 1] + constant Values are merged in order, so if an index appears in both `indices[m][i]` and `indices[n][j]` for `(m,i) < (n,j)` the slice `data[n][j]` will appear in the merged result. If you do not need this guarantee, ParallelDynamicStitch might perform better on some devices. For example: ```python indices[0] = 6 indices[1] = [4, 1] indices[2] = [[5, 2], [0, 3]] data[0] = [61, 62] data[1] = [[41, 42], [11, 12]] data[2] = [[[51, 52], [21, 22]], [[1, 2], [31, 32]]] merged = [[1, 2], [11, 12], [21, 22], [31, 32], [41, 42], [51, 52], [61, 62]] ``` This method can be used to merge partitions created by `dynamic_partition` as illustrated on the following example: ```python # Apply function (increments x_i) on elements for which a certain condition # apply (x_i != -1 in this example). x=tf.constant([0.1, -1., 5.2, 4.3, -1., 7.4]) condition_mask=tf.not_equal(x,tf.constant(-1.)) partitioned_data = tf.dynamic_partition( x, tf.cast(condition_mask, tf.int32) , 2) partitioned_data[1] = partitioned_data[1] + 1.0 condition_indices = tf.dynamic_partition( tf.range(tf.shape(x)[0]), tf.cast(condition_mask, tf.int32) , 2) x = tf.dynamic_stitch(condition_indices, partitioned_data) # Here x=[1.1, -1., 6.2, 5.3, -1, 8.4], the -1. values remain # unchanged. ```
Args: indices: A list of at least 1 `Tensor` objects with type `int32`. data: A list with the same length as `indices` of `Tensor` objects with the same type. name: A name for the operation (optional). Returns: A `Tensor`. Has the same type as `data`. DynamicStitchNrrGrrr.DExpected list for 'indices' argument to 'dynamic_stitch' Op, not %r.AExpected list for 'data' argument to 'dynamic_stitch' Op, not %r.fList argument 'data' to 'dynamic_stitch' Op with length %d must match length %d of argument 'indices'.Nr)#r0rr1r2rrrrrrr_dispatcher_for_dynamic_stitchrdynamic_stitch_eager_fallbackrr|rrr rrrrryrzr{lenr4r5rIrJrrgrKrL rrr.r6r7rNr_attr_Nr8r9r:rOrPs r;rrsZ    0h..0$ #\\ 11 otWd4g n,- $&Gn$ n GdE] +  ')0 1 22 L' D4- (  ')- . // Y'  6 TG    )::t$@Aq#x QK' ""$3$$S)30B0B30G HF::L vw8 (' .s  & &- ##At,,  # #    . D$ $(g  & * 4d..  # #  z " "" BW4d K g  ..<< <  < Z    "d7DI Gi,,::: n  sPE4/I4F;F""F;:F;?G'G''I=AIIAJ/-J/zraw_ops.DynamicStitchcXt|ttfstd|zt |}t|ttfstd|zt ||k7rt dt ||fzt jt||g\}}tj|tj}t|t|z}d|d|f}t jdd||||}t jrt jd ||||\}|S) Nr r r rrs DynamicStitchrr ryrzr{r|rrrIrrconvert_n_to_tensorrrrrJrL rrr.r@rrrPrOrNs r;rrqs/ GdE] +  ')0 1 22 L' D4- (  ')- . // Y'  6 TG  11$t*c2F-'4  $ $Wgmm <'gd+, #w '&   -q#)s ?' ""$ vw8 (' .r=c tjxstj}|j}|jr t dt |t tfstd|z|Dcgc]}tj|d}}|g}t |t tfstd|z|D cgc]} tj| d}} |d}tj|d}|d }tj|d }|d }tj|d }tjd |||||| \} } } } | dd} tj r{d| j#dd| j#dd| j%dd | j#d d | j#d f }| j&}tj(d ||| | \} | Scc}wcc} w)auA queue that produces elements in first-in first-out order. Args: component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a value. shapes: An optional list of shapes (each a `tf.TensorShape` or list of `ints`). Defaults to `[]`. The shape of each component in a value. The length of this attr must be either 0 or the same as the length of component_types. If the length of this attr is 0, the shapes of queue elements are not constrained, and only one element may be dequeued at a time. capacity: An optional `int`. Defaults to `-1`. The upper bound on the number of elements in this queue. Negative numbers mean no limit. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this queue will be shared under the given name across multiple sessions. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Ffifo_queue op does not support eager execution. Arg 'handle' is a ref.zHExpected list for 'component_types' argument to 'fifo_queue' Op, not %r.rtNz?Expected list for 'shapes' argument to 'fifo_queue' Op, not %r.rurmrornrprq FIFOQueuerwrxrs r; fifo_queuers2    0h..0$ #\\ _ `` OdE] 3  #%4 5 66JYY2X'',=>Y/Y ^ F FT5M *  #%+ , --9? ?"H  H - ?& ? H   x 4(I ;7)K!!+}=+'88_V&)!,49!QX QK' ""$.?!@(ll8$j +[ll;'ll=) +F ::L \674 (' .AZ @rzraw_ops.FIFOQueuectd)Nrr?rs r;fifo_queue_eager_fallbackr]^^r=ctjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |g}t|ttfst!d |z|D cgc]} t#j&| d}} |d }t#j(|d}|d }t#j*|d}|d }t#j*|d}t-j.d|||||| \} } } }|dd}t#j0r{d| j3dd| j3dd| j5dd| j3dd| j3df }| j6}t#j8d||||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} wcc} w)aoA queue that produces elements in first-in first-out order. Args: component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a value. shapes: An optional list of shapes (each a `tf.TensorShape` or list of `ints`). Defaults to `[]`. The shape of each component in a value. The length of this attr must be either 0 or the same as the length of component_types. If the length of this attr is 0, the shapes of queue elements are not constrained, and only one element may be dequeued at a time. capacity: An optional `int`. Defaults to `-1`. The upper bound on the number of elements in this queue. Negative numbers mean no limit. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this queue will be shared under the given name across multiple sessions. name: A name for the operation (optional). Returns: A `Tensor` of type `resource`. FIFOQueueV2rtrurorprqNrKExpected list for 'component_types' argument to 'fifo_queue_v2' Op, not %r.BExpected list for 'shapes' argument to 'fifo_queue_v2' Op, not %r.rmrnrw)r0rr1r2rrrrrrrfifo_queue_v2_eager_fallbackrryrzr{r|rIrfr}r~rr4r5rJrrrKrLrtrurorprqr.r6r7rNrrrr8r9r:rOrPs r; fifo_queue_v2r&s2    0h..0$ #\\  11 mT#4o&*h Y{$gn OdE] 3  &(7 8 99JYY2X'',=>Y/Y ^ F FT5M *  &(. / 009? ?"H  H - ?& ? H   x 4(I ;7)K!!+}=+'88v (I#.T;!QX QK' ""$.?!@(ll8$j +[ll;'ll=) +F ::L |VW6 (' ._  & &- ##At,,  # #    ))&8;TtMM  # #   Z @<"G> I5I:>IH,,II II21I2zraw_ops.FIFOQueueV2c t|ttfstd|z|Dcgc]}t j |d}}|g}t|ttfstd|z|Dcgc]}t j |d}}|d}t j|d}|d}t j|d}|d}t j|d }g} d|d|d|d|d |f } t jd d | | || } t jrt jd | | | | \} | Scc}wcc}w)Nr"rtr#rurmrornrprqs FIFOQueueV2rrr! ryrzr{r|rIrfr}r~rrrJrL rtrurorprqr.r@rrrPrOrNs r;r$r$-su OdE] 3  &(7 8 99JYY2X'',=>Y/Y ^ F FT5M *  &(. / 009? ?"H  H - ?& ? H   x 4(I ;7)K!!+}=+, &* KM; @&   ^Q|#)s ?' ""$ |VW6 (' .5Z @ D<4EresourcecXtjxstj}|j}|jr t dt j d||\}}}}|dd}tjr&d}|j} tjd| |||\}|S)zDeprecated. Do not use. Args: resource: A `Tensor` of type `resource`. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Ffake_queue op does not support eager execution. Arg 'handle' is a ref. FakeQueue)r,r.NrGrH) r,r.r6r7r8r9r:rNrOrPs r; fake_queuer0Os    0h..0$ #\\ _ ``'88hT3!QX QK' ""$ F::L \674 (' .r=zraw_ops.FakeQueuectd)Nr.r?)r,r.r@s r;fake_queue_eager_fallbackr2lrr=) TV_GetSessionHandle_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rvaluec~tjxstj}|j}|jr t j |d||}|Stjd||\}}}}|dd}tj r7d|j#df} |j$} tj&d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||S#t j$rYwxYw)zStore the input tensor in the state of the current session. Args: value: A `Tensor`. The tensor to be stored. name: A name for the operation (optional). Returns: A `Tensor` of type `string`. GetSessionHandleNrr4r.r)r0rr1r2rrrrrrr!get_session_handle_eager_fallbackrr4r5rIrJrgrKrL r4r.r6r7rNrr8r9r:rOrPs r;get_session_handler:qs8    0h..0$ #\\ 11  $/g n(88%d4!QX QK' ""$3%%c* +F::L L&'; (' .'  & &- ##At,,  # #    . d&&  # #   0C D C;;DD D&&D<;D<zraw_ops.GetSessionHandlectj|g|g\}\}|g}d|f}tjdd||||}tjrtjd||||\}|S)NrsGetSessionHandlerrr6rIrrrJrLr4r.r@rrPrOrNs r;r8r8s|55ugsBG'8E, >&   0!L#)s ?' ""$ L&'; (' .r=) TV_GetSessionHandleV2_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc~tjxstj}|j}|jr t j |d||}|Stjd||\}}}}|dd}tj r7d|j#df} |j$} tj&d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||S#t j$rYwxYw)zStore the input tensor in the state of the current session. Args: value: A `Tensor`. The tensor to be stored. name: A name for the operation (optional). Returns: A `Tensor` of type `resource`. GetSessionHandleV2Nrr7r)r0rr1r2rrrrrrr$get_session_handle_v2_eager_fallbackrr4r5rIrJrgrKrLr9s r;get_session_handle_v2rCs8    0h..0$ #\\ 11 "D%1g n(88E6!QX QK' ""$3%%c* +F::L lFG= (' .'  & &- ##At,,  # #    1 d&&  # #   r;zraw_ops.GetSessionHandleV2ctj|g|g\}\}|g}d|f}tjdd||||}tjrtjd||||\}|S)NrsGetSessionHandleV2rrrAr=r>s r;rBrBs|55ugsBG'8E, >&   2Al#)s ?' ""$ lFG= (' .r=) TV_GetSessionTensor_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rctjxstj}|j}|jr t j |d||d|}|Stj|d}tj d|||\}}}} | dd}tj"r7d|j%df} |j&} tj(d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||S#t j$rYwxYw)a6Get the value of the tensor specified by its handle. Args: handle: A `Tensor` of type `string`. The handle for a tensor stored in the session state. dtype: A `tf.DType`. The type of the output value. name: A name for the operation (optional). Returns: A `Tensor` of type `dtype`. GetSessionTensorr`Nr`r.r@)r'r`r.)r0rr1r2rrrrrrr!get_session_tensor_eager_fallbackrrIrfr4r5rJrgrKrL) r'r`r.r6r7rNrr8r9r:rOrPs r;get_session_tensorrJsT    0h..0$ #\\ 11  $@g n   UG ,%'886TC!QX QK' ""$s))'2 3F::L L&'; (' .)  & &- ##At,,  # #    . Dd44  # #   0C&&D-9DD-,D-1EEEzraw_ops.GetSessionTensorctj|d}tj|tj }|g}d|f}tj dd||||}tjrtjd||||\}|S)Nr`sGetSessionTensorrrrG) rIrfrrrrrrJrL)r'r`r.r@rPrOrNs r;rIrI s   UG ,%  ! !&'.. 9&, U &   0!L#)s ?' ""$ L&'; (' .r= memory_limitcbtjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYEwxYwcc} w) Op removes all elements in the underlying container. Args: dtypes: A list of `tf.DTypes`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: The created Operation. MapClearrorMrrprqNrorMrrprqr.r@>Expected list for 'dtypes' argument to 'map_clear' Op, not %r.rrnrrorMrprqr.)r0rr1r2rrrrrrrmap_clear_eager_fallbackrryrzr{r|rIrfr~rr4r5rrorMrprqr.r6r7rNrrr8r9r:s r; map_clearrVs    0h..0$ #\\  11 j$ Hnh Y gn FT5M *  "$* + ,,8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'886H!- +$8!QX *A  & &- ##At,,  # #    %,v;TtMM  # #    ?6"D5 F,5E<E##E<;E<FF)(F)zraw_ops.MapClearc t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd d|| || } d} | Scc}w) NrRrrrorMrnrprqsMapClearr ryrzr{r|rIrfr~rr rrorMrprqr.r@rrPrOrNs r;rTrTVs FT5M *  "$* + ,,8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   [!L#)s ?' ' .' ?Cctjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | dd}t#j.r{d| j1dd| j1dd| j3dd| j3dd| j3df }| j4}t#j6d||||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} w) Op returns the number of incomplete elements in the underlying container. Args: dtypes: A list of `tf.DTypes`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. MapIncompleteSizerorMrrprqNrQHExpected list for 'dtypes' argument to 'map_incomplete_size' Op, not %r.rrnrS)r0rr1r2rrrrrrr"map_incomplete_size_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLrrorMrprqr.r6r7rNrrr8r9r:rOrPs r;map_incomplete_sizerbqsh    0h..0$ #\\  11 !4X~h Y gn FT5M *  ,.4 5 668> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88FX*6))44A!QX QK' ""$#++J7/ll8$k3<< 3LS\\-8:F::L \67< (' .U  & &- ##At,,  # #    /,v;TtMM  # #    ?6"G I H G;;HHH**IIzraw_ops.MapIncompleteSizec  t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd d || || } t jrt jd || | | \} | Scc}w) Nr_rrrorMrnrprqsMapIncompleteSizerrr^ ryrzr{r|rIrfr~rrrJrLrZs r;r`r`s4 FT5M *  ,.4 5 668> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   11\#)s ?' ""$ \67< (' .- ?D keyctjxstj}|j} | jr& t j |d|||d|d|d|d|d|} | St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||||| \} } }}|dd} | s|St#j.r{d|j1dd|j1dd|j3dd|j3dd|j3df }|j4}t#j6d||| | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t||||||||| S#t j$rYwxYwcc} w) aOp peeks at the values at the specified key. If the underlying container does not contain this key this op will block until it does. Args: key: A `Tensor` of type `int64`. indices: A `Tensor` of type `int32`. dtypes: A list of `tf.DTypes` that has length `>= 1`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `dtypes`. MapPeekrorMrrprqNrQ=Expected list for 'dtypes' argument to 'map_peek' Op, not %r.rrnrgrrrorMrprqr.)r0rr1r2rrrrrrrmap_peek_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLrgrrrorMrprqr.r6r7rNrrr8r9r:rOrPs r;map_peekrnsr&    0h..0$ #\\  11 isGZ h =+/gn FT5M *  !#) * ++8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88sGFX , *7!QX QK'  J ""$#++J7/ll8$k3<< 3LS\\-8:F::L <2 .Y  & &- ##At,,  # #    $ w 9+  # #    ?6$G I H$G??HHH00IIzraw_ops.MapPeekc t|ttfstd|z|D cgc]} t j | d}} |d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}tj|tj}tj|tj}||g} d|d|d|d|d|f } t jd t|| | || } t jrt j d | | | | Scc} w) NrjrrrorMrnrprqsMapPeekrriryrzr{r|rIrfr~rrrrint64rrrrJrL rgrrrorMrprqr.r@rrPrOrNs r;rlrl"s` FT5M *  !#) * ++8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+ sGMM2#  " "7GMM :'w, .,+y- >&   ZV\#)s ?' ""$ <2 ./ ?Ectjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | dd}t#j.r{d| j1dd| j1dd| j3dd| j3dd| j3df }| j4}t#j6d||||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} w) Op returns the number of elements in the underlying container. Args: dtypes: A list of `tf.DTypes`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. MapSizerorMrrprqNrQ=Expected list for 'dtypes' argument to 'map_size' Op, not %r.rrnrS)r0rr1r2rrrrrrrmap_size_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLras r;map_sizerzAsf    0h..0$ #\\  11 iz8^h Y gn FT5M *  !#) * ++8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88&8 , *7!QX QK' ""$#++J7/ll8$k3<< 3LS\\-8:F::L <2 (' .U  & &- ##At,,  # #    $,v;TtMM  # #    ?rczraw_ops.MapSizec  t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd d || || } t jrt jd || | | \} | Scc}w) NrxrrrorMrnrprqsMapSizerrrwrerZs r;ryrys3 FT5M *  !#) * ++8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   Z= 0`. Defaults to `0`. Maximum number of elements in the Staging Area. If > 0, inserts on the container will block when the capacity is reached. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. It is necessary to match this name to the matching Unstage Op. name: A name for the operation (optional). Returns: The created Operation. MapStagerorMrrprqNrQ>Expected list for 'dtypes' argument to 'map_stage' Op, not %r.rrn rgrrrrorMrprqr.)r0rr1r2rrrrrrrmap_stage_eager_fallbackrryrzr{r|rIrfr~rr4r5rgrrrrorMrprqr.r6r7rNrrr8r9r:s r; map_stagers.    0h..0$ #\\  11 j$Wfj( h =+/gn FT5M *  "$* + ,,8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88WVF%L'[tM!QX *C  & &- ##At,,  # #    % w 9+  # #    ?6%D;F5;FE))FFFF21F2zraw_ops.MapStagec t|ttfstd|z|D cgc]} t j | d}} |d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}t j|| \} }tj|tj}tj|tj}||gt|z} d|d|d|d | d|d|f } t jd d| | | | }d}|Scc} w) Nr~rrrorMrnrprq fake_dtypessMapStagerryrzr{r|rIrfr~rconvert_to_mixed_eager_tensorsrrrrrrrrgrrrrorMrprqr.r@r_attr_fake_dtypesrPrOrNs r;rrsg FT5M *  "$* + ,,8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+&EEfcRV sGMM2#  " "7GMM :'w$v,., .,-*K &   [!L#)s ?' ' ./ ?Ectjxstj}|j} | jr& t j |d|||d|d|d|d|d|} | St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||||| \} } }}|dd} | s|St#j.r{d|j1dd|j1dd|j3dd|j3dd|j3df }|j4}t#j6d||| | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t||||||||| S#t j$rYwxYwcc} w) Op removes and returns the values associated with the key from the underlying container. If the underlying container does not contain this key, the op will block until it does. Args: key: A `Tensor` of type `int64`. indices: A `Tensor` of type `int32`. dtypes: A list of `tf.DTypes` that has length `>= 1`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `dtypes`. MapUnstagerorMrrprqNrQ@Expected list for 'dtypes' argument to 'map_unstage' Op, not %r.rrnrk)r0rr1r2rrrrrrrmap_unstage_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLrms r; map_unstager ss&    0h..0$ #\\  11 lD#w H h =+/gn FT5M *  $&, - ..8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88#wv'l ){O!QX QK'  J ""$#++J7/ll8$k3<< 3LS\\-8:F::L lFG5 .Y  & &- ##At,,  # #    ' w 9+  # #    ?rozraw_ops.MapUnstagec t|ttfstd|z|D cgc]} t j | d}} |d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}tj|tj}tj|tj}||g} d|d|d|d|d|f } t jd t|| | || } t jrt j d | | | | Scc} w) NrrrrorMrnrprqs MapUnstagerrrqrss r;rrVs` FT5M *  $&, - ..8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+ sGMM2#  " "7GMM :'w, .,+y- >&   ]CK #)s ?' ""$ lFG5 ./ ?rtMapUnstageNoKeyctjxstj}|j}|jr: t j |d||d|d|d|d|d|} t j| } | St|t t"fst%d |z|D cgc]} t'j(| d}} |d }t'j*|d}|d }t'j*|d}|d }t'j,|d}|d }t'j,|d}t/j0d||||||| \} } } }|dd} t'j2r{d| j5dd| j5dd| j7dd| j7dd| j7df }| j8}t'j:d||| | dd | d dgz} t j| } | S#tj$r } tj| |Yd} ~ nd} ~ wtj$rYnwxYw t||||||||S#tj$rYwxYwcc} w)aOp removes and returns a random (key, value) from the underlying container. If the underlying container does not contain elements, the op will block until it does. Args: indices: A `Tensor` of type `int32`. dtypes: A list of `tf.DTypes` that has length `>= 1`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (key, values). key: A `Tensor` of type `int64`. values: A list of `Tensor` objects of type `dtypes`. rrorMrrprqNrQGExpected list for 'dtypes' argument to 'map_unstage_no_key' Op, not %r.rrnrrrorMrprqr.r)r0rr1r2rr_MapUnstageNoKeyOutputrrrrrr!map_unstage_no_key_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLrrrorMrprqr.r6r7rNrrr8r9r:rOrPs r;map_unstage_no_keyrys*    0h..0$ #\\  11 w H h =+/g',,W5g n FT5M *  +-3 4 558> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'8876H(4 '2?!QX QK' ""$#++J7/ll8$k3<< 3LS\\-8:F::L <: BQK712;- '' " ( ( 1' .Y  & &- ##At,,  # #    . H<9+  # #    ?68H!I9IH//II II65I6zraw_ops.MapUnstageNoKeyc t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}tj|tj}|g} d|d|d|d|d|f } t jd t|d z| | || } t jrt jd | | | | dd | d dgz} t j#| } | Scc}w) NrrrrorMrnrprqsMapUnstageNoKeyrrr)ryrzr{r|rIrfr~rrrrrrrrJrLrr rrrorMrprqr.r@rrPrOrNs r;rrsx FT5M *  +-3 4 558> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+  " "7GMM :', .,+y- >&   /Vq$0C"& ('""$ <: BQK712;- '' " ( ( 1' .3 ?Ecbtjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYEwxYwcc} w) rOOrderedMapClearrorMrrprqNrQFExpected list for 'dtypes' argument to 'ordered_map_clear' Op, not %r.rrnrS)r0rr1r2rrrrrrr ordered_map_clear_eager_fallbackrryrzr{r|rIrfr~rr4r5rUs r;ordered_map_clearrs    0h..0$ #\\  11 z8^h Y gn FT5M *  *,2 3 448> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88&8(4 '2?!QX *A  & &- ##At,,  # #    -,v;TtMM  # #    ?rWzraw_ops.OrderedMapClearc t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd d|| || } d} | Scc}w) NrrrrorMrnrprqsOrderedMapClearrrYrZs r;rr%s FT5M *  *,2 3 448> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   /<#)s ?' ' .' ?r[ctjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | dd}t#j.r{d| j1dd| j1dd| j3dd| j3dd| j3df }| j4}t#j6d||||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} w) r]OrderedMapIncompleteSizerorMrrprqNrQPExpected list for 'dtypes' argument to 'ordered_map_incomplete_size' Op, not %r.rrnrS)r0rr1r2rrrrrrr*ordered_map_incomplete_size_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLras r;ordered_map_incomplete_sizer@sj    0h..0$ #\\  11 ($ H h =+/gn FT5M *  46< = >>8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88"6H1=.70;$ H!QX QK' ""$#++J7/ll8$k3<< 3LS\\-8:F::L "L&'C (' .W  & &- ##At,,  # #    7,v;TtMM  # #    ?rcz raw_ops.OrderedMapIncompleteSizec  t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd d || || } t jrt jd || | | \} | Scc}w) NrrrrorMrnrprqsOrderedMapIncompleteSizerrrrerZs r;rrs8 FT5M *  46< = >>8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   8!$0C"& ('""$ "L&'C (' ./ ?rfctjxstj}|j} | jr& t j |d|||d|d|d|d|d|} | St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||||| \} } }}|dd} | s|St#j.r{d|j1dd|j1dd|j3dd|j3dd|j3df }|j4}t#j6d||| | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t||||||||| S#t j$rYwxYwcc} w) aOp peeks at the values at the specified key. If the underlying container does not contain this key this op will block until it does. This Op is optimized for performance. Args: key: A `Tensor` of type `int64`. indices: A `Tensor` of type `int32`. dtypes: A list of `tf.DTypes` that has length `>= 1`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `dtypes`. OrderedMapPeekrorMrrprqNrQEExpected list for 'dtypes' argument to 'ordered_map_peek' Op, not %r.rrnrk)r0rr1r2rrrrrrrordered_map_peek_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLrms r;ordered_map_peekrst(    0h..0$ #\\  11 c7J h =+/gn FT5M *  )+1 2 338> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88c76#+,$-;# %!QX QK'  J ""$#++J7/ll8$k3<< 3LS\\-8:F::L ,9 .[  & &- ##At,,  # #    , w 9+  # #    ?rozraw_ops.OrderedMapPeekc t|ttfstd|z|D cgc]} t j | d}} |d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}tj|tj}tj|tj}||g} d|d|d|d|d|f } t jd t|| | || } t jrt j d | | | | Scc} w) NrrrrorMrnrprqsOrderedMapPeekrrrqrss r;rrsd FT5M *  )+1 2 338> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+ sGMM2#  " "7GMM :'w, .,+y- >&   .F $0C"& ('""$ ,9 .1 ?rtctjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | dd}t#j.r{d| j1dd| j1dd| j3dd| j3dd| j3df }| j4}t#j6d||||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} w) rvOrderedMapSizerorMrrprqNrQEExpected list for 'dtypes' argument to 'ordered_map_size' Op, not %r.rrnrS)r0rr1r2rrrrrrrordered_map_size_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLras r;ordered_map_sizer sg    0h..0$ #\\  11 j(Nh Y gn FT5M *  )+1 2 338> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88('3y&1>!QX QK' ""$#++J7/ll8$k3<< 3LS\\-8:F::L ,9 (' .U  & &- ##At,,  # #    ,,v;TtMM  # #    ?rczraw_ops.OrderedMapSizec  t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd d || || } t jrt jd || | | \} | Scc}w) NrrrrorMrnrprqsOrderedMapSizerrrrerZs r;rr[ s4 FT5M *  )+1 2 338> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   .,#)s ?' ""$ ,9 (' .- ?rfc ttjxstj} | j} | jr' t j | d||||d|d|d|d|d|} | St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d||||||||| \}}}}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t||||||||||  S#t j$rYKwxYwcc} w) aStage (key, values) in the underlying container which behaves like a ordered associative container. Elements are ordered by key. Args: key: A `Tensor` of type `int64`. int64 indices: A `Tensor` of type `int32`. values: A list of `Tensor` objects. a list of tensors dtypes A list of data types that inserted values should adhere to. dtypes: A list of `tf.DTypes`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. Maximum number of elements in the Staging Area. If > 0, inserts on the container will block when the capacity is reached. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. It is necessary to match this name to the matching Unstage Op. name: A name for the operation (optional). Returns: The created Operation. OrderedMapStagerorMrrprqNrQFExpected list for 'dtypes' argument to 'ordered_map_stage' Op, not %r.rrnr)r0rr1r2rrrrrrr ordered_map_stage_eager_fallbackrryrzr{r|rIrfr~rr4r5rs r;ordered_map_stagery s2    0h..0$ #\\  11 sGVZ.,&+=+/gn FT5M *  *,2 3 448> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88sGF"(8(4 '2 ?!QX *E  & &- ##At,,  # #    - w 9+  # #    ?rzraw_ops.OrderedMapStagec t|ttfstd|z|D cgc]} t j | d}} |d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}t j|| \} }tj|tj}tj|tj}||gt|z} d|d|d|d | d|d|f } t jd d| | | | }d}|Scc} w) NrrrrorMrnrprqrsOrderedMapStagerrrs r;rr sh FT5M *  *,2 3 448> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+&EEfcRV sGMM2#  " "7GMM :'w$v,., .,-*K &   /<#)s ?' ' ./ ?rctjxstj}|j} | jr& t j |d|||d|d|d|d|d|} | St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||||| \} } }}|dd} | s|St#j.r{d|j1dd|j1dd|j3dd|j3dd|j3df }|j4}t#j6d||| | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t||||||||| S#t j$rYwxYwcc} w) rOrderedMapUnstagerorMrrprqNrQHExpected list for 'dtypes' argument to 'ordered_map_unstage' Op, not %r.rrnrk)r0rr1r2rrrrrrr"ordered_map_unstage_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLrms r;ordered_map_unstager st&    0h..0$ #\\  11 !4gz8 h =+/gn FT5M *  ,.4 5 668> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88gf&.\'0k"& (!QX QK'  J ""$#++J7/ll8$k3<< 3LS\\-8:F::L \67< .[  & &- ##At,,  # #    / w 9+  # #    ?rozraw_ops.OrderedMapUnstagec t|ttfstd|z|D cgc]} t j | d}} |d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}tj|tj}tj|tj}||g} d|d|d|d|d|f } t jd t|| | || } t jrt j d | | | | Scc} w) NrrrrorMrnrprqsOrderedMapUnstagerrrqrss r;rr. sd FT5M *  ,.4 5 668> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+ sGMM2#  " "7GMM :'w, .,+y- >&   13v;$0C"& ('""$ \67< .1 ?rtOrderedMapUnstageNoKeyctjxstj}|j}|jr: t j |d||d|d|d|d|d|} t j| } | St|t t"fst%d |z|D cgc]} t'j(| d}} |d }t'j*|d}|d }t'j*|d}|d }t'j,|d}|d }t'j,|d}t/j0d||||||| \} } } }|dd} t'j2r{d| j5dd| j5dd| j7dd| j7dd| j7df }| j8}t'j:d||| | dd | d dgz} t j| } | S#tj$r } tj| |Yd} ~ nd} ~ wtj$rYnwxYw t||||||||S#tj$rYwxYwcc} w)aOp removes and returns the (key, value) element with the smallest key from the underlying container. If the underlying container does not contain elements, the op will block until it does. Args: indices: A `Tensor` of type `int32`. dtypes: A list of `tf.DTypes` that has length `>= 1`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (key, values). key: A `Tensor` of type `int64`. values: A list of `Tensor` objects of type `dtypes`. rrorMrrprqNrQOExpected list for 'dtypes' argument to 'ordered_map_unstage_no_key' Op, not %r.rrnrr)r0rr1r2rr_OrderedMapUnstageNoKeyOutputrrrrrr)ordered_map_unstage_no_key_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLrs r;ordered_map_unstage_no_keyrR s*    0h..0$ #\\  11 &gz8 h =+/g.33G >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88 '&+3/;,5.9 F!QX QK' ""$#++J7/ll8$k3<< 3LS\\-8:F::L  ,A BQK712;- '' ) / / 8' .]  & &- ##At,,  # #    6 H<9+  # #    ?rzraw_ops.OrderedMapUnstageNoKeyc t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}tj|tj}|g} d|d|d|d|d|f } t jd t|d z| | || } t jrt jd | | | | dd | d dgz} t j#| } | Scc}w) NrrrrorMrnrprqsOrderedMapUnstageNoKeyrrr)ryrzr{r|rIrfr~rrrrrrrrJrLrrrs r;rr sy FT5M *  35; < ==8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+  " "7GMM :', .,+y- >&   6F a$0C"& ('""$  ,A BQK712;- '' ) / / 8' .3 ?rc tjxstj}|j}|jr t dt |t tfstd|z|Dcgc]}tj|d}}|g}t |t tfstd|z|D cgc]} tj| d}} |d}tj|d}|d }tj|d }|d }tj|d }tjd |||||| \} } } } | dd} tj r{d| j#dd| j#dd| j%dd | j#d d | j#d f }| j&}tj(d ||| | \} | Scc}wcc} w)aA queue that produces elements in first-in first-out order. Variable-size shapes are allowed by setting the corresponding shape dimensions to 0 in the shape attr. In this case DequeueMany will pad up to the maximum size of any given element in the minibatch. See below for details. Args: component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a value. shapes: An optional list of shapes (each a `tf.TensorShape` or list of `ints`). Defaults to `[]`. The shape of each component in a value. The length of this attr must be either 0 or the same as the length of component_types. Shapes of fixed rank but variable size are allowed by setting any shape dimension to -1. In this case, the inputs' shape may vary along the given dimension, and DequeueMany will pad the given dimension with zeros up to the maximum shape of all elements in the given batch. If the length of this attr is 0, different queue elements may have different ranks and shapes, but only one element may be dequeued at a time. capacity: An optional `int`. Defaults to `-1`. The upper bound on the number of elements in this queue. Negative numbers mean no limit. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this queue will be shared under the given name across multiple sessions. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Npadding_fifo_queue op does not support eager execution. Arg 'handle' is a ref.zPExpected list for 'component_types' argument to 'padding_fifo_queue' Op, not %r.rtNzGExpected list for 'shapes' argument to 'padding_fifo_queue' Op, not %r.rurmrornrprqPaddingFIFOQueuerwrxrs r;padding_fifo_queuer sB    0h..0$ #\\ g hh OdE] 3  +-< = >>JYY2X'',=>Y/Y ^ F FT5M *  +-3 4 559? ?"H  H - ?& ? H   x 4(I ;7)K!!+}=+'88OF%-(3$@!QX QK' ""$.?!@(ll8$j +[ll;'ll=) +F ::L L&'; (' .AZ @rzraw_ops.PaddingFIFOQueuectd)Nrr?rs r;!padding_fifo_queue_eager_fallbackr rr=ctjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |g}t|ttfst!d |z|D cgc]} t#j&| d}} |d }t#j(|d}|d }t#j*|d}|d }t#j*|d}t-j.d|||||| \} } } }|dd}t#j0r{d| j3dd| j3dd| j5dd| j3dd| j3df }| j6}t#j8d||||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} wcc} w)aA queue that produces elements in first-in first-out order. Variable-size shapes are allowed by setting the corresponding shape dimensions to 0 in the shape attr. In this case DequeueMany will pad up to the maximum size of any given element in the minibatch. See below for details. Args: component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a value. shapes: An optional list of shapes (each a `tf.TensorShape` or list of `ints`). Defaults to `[]`. The shape of each component in a value. The length of this attr must be either 0 or the same as the length of component_types. Shapes of fixed rank but variable size are allowed by setting any shape dimension to -1. In this case, the inputs' shape may vary along the given dimension, and DequeueMany will pad the given dimension with zeros up to the maximum shape of all elements in the given batch. If the length of this attr is 0, different queue elements may have different ranks and shapes, but only one element may be dequeued at a time. capacity: An optional `int`. Defaults to `-1`. The upper bound on the number of elements in this queue. Negative numbers mean no limit. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this queue will be shared under the given name across multiple sessions. name: A name for the operation (optional). Returns: A `Tensor` of type `resource`. PaddingFIFOQueueV2rtrurorprqNrSExpected list for 'component_types' argument to 'padding_fifo_queue_v2' Op, not %r.JExpected list for 'shapes' argument to 'padding_fifo_queue_v2' Op, not %r.rmrnrw)r0rr1r2rrrrrrr$padding_fifo_queue_v2_eager_fallbackrryrzr{r|rIrfr}r~rr4r5rJrrrKrLr%s r;padding_fifo_queue_v2r sB    0h..0$ #\\  11 "D*;_&*h Y{$gn OdE] 3  .0? @ AAJYY2X'',=>Y/Y ^ F FT5M *  .06 7 889? ?"H  H - ?& ? H   x 4(I ;7)K!!+}=+'88of'/9*5DB!QX QK' ""$.?!@(ll8$j +[ll;'ll=) +F ::L lFG= (' ._  & &- ##At,,  # #    1)&8;TtMM  # #   Z @r'zraw_ops.PaddingFIFOQueueV2c t|ttfstd|z|Dcgc]}t j |d}}|g}t|ttfstd|z|Dcgc]}t j |d}}|d}t j|d}|d}t j|d}|d}t j|d }g} d|d|d|d|d |f } t jd d | | || } t jrt jd | | | | \} | Scc}wcc}w)NrrtrrurmrornrprqsPaddingFIFOQueueV2rrrr)r*s r;rru sy OdE] 3  .0? @ AAJYY2X'',=>Y/Y ^ F FT5M *  .06 7 889? ?"H  H - ?& ? H   x 4(I ;7)K!!+}=+, &* KM; @&   2Al#)s ?' ""$ lFG= (' .5Z @r+) TV_ParallelDynamicStitch_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rctjxstj}|j}|jr t j |d|||}|St|ttfst!d|zt#|}t|ttfst!d|zt#||k7rt%dt#||fzt'j(d|||\}}} } | dd}t+j,rHd| j/dd | j1d f} | j2} t+j4d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||S#t j$rYowxYw) a Interleave the values from the `data` tensors into a single tensor. Builds a merged tensor such that ```python merged[indices[m][i, ..., j], ...] = data[m][i, ..., j, ...] ``` For example, if each `indices[m]` is scalar or vector, we have ```python # Scalar indices: merged[indices[m], ...] = data[m][...] # Vector indices: merged[indices[m][i], ...] = data[m][i, ...] ``` Each `data[i].shape` must start with the corresponding `indices[i].shape`, and the rest of `data[i].shape` must be constant w.r.t. `i`. That is, we must have `data[i].shape = indices[i].shape + constant`. In terms of this `constant`, the output shape is merged.shape = [max(indices)] + constant Values may be merged in parallel, so if an index appears in both `indices[m][i]` and `indices[n][j]`, the result may be invalid. This differs from the normal DynamicStitch operator that defines the behavior in that case. For example: ```python indices[0] = 6 indices[1] = [4, 1] indices[2] = [[5, 2], [0, 3]] data[0] = [61, 62] data[1] = [[41, 42], [11, 12]] data[2] = [[[51, 52], [21, 22]], [[1, 2], [31, 32]]] merged = [[1, 2], [11, 12], [21, 22], [31, 32], [41, 42], [51, 52], [61, 62]] ``` This method can be used to merge partitions created by `dynamic_partition` as illustrated on the following example: ```python # Apply function (increments x_i) on elements for which a certain condition # apply (x_i != -1 in this example). x=tf.constant([0.1, -1., 5.2, 4.3, -1., 7.4]) condition_mask=tf.not_equal(x,tf.constant(-1.)) partitioned_data = tf.dynamic_partition( x, tf.cast(condition_mask, tf.int32) , 2) partitioned_data[1] = partitioned_data[1] + 1.0 condition_indices = tf.dynamic_partition( tf.range(tf.shape(x)[0]), tf.cast(condition_mask, tf.int32) , 2) x = tf.dynamic_stitch(condition_indices, partitioned_data) # Here x=[1.1, -1., 6.2, 5.3, -1, 8.4], the -1. values remain # unchanged. ```
Args: indices: A list of at least 1 `Tensor` objects with type `int32`. data: A list with the same length as `indices` of `Tensor` objects with the same type. name: A name for the operation (optional). Returns: A `Tensor`. Has the same type as `data`. ParallelDynamicStitchNrMExpected list for 'indices' argument to 'parallel_dynamic_stitch' Op, not %r.JExpected list for 'data' argument to 'parallel_dynamic_stitch' Op, not %r.oList argument 'data' to 'parallel_dynamic_stitch' Op with length %d must match length %d of argument 'indices'.r rr)r0rr1r2rrrrrrr¶llel_dynamic_stitch_eager_fallbackrryrzr{r|rrr4r5rIrJrrgrKrLrs r;parallel_dynamic_stitchr sR    0h..0$ #\\ 11 %tWd    0h..0$ #\\ c dd FT5M *  ')/ 0 119? ?"H  H - ?& ?O OdE] 3  ')8 9 ::JYY2X'',=>Y/Y H   x 4(I ;7)K!!+}=+'88"*i%0t=!QX QK' ""$.?!@(ll8$j +[ll;'ll=) +F ::L vw8 (' .A @Zrzraw_ops.PriorityQueuectd)Nrr?)rurtrorprqr.r@s r;priority_queue_eager_fallbackrx sabbr=ctjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |g}t|ttfst!d |z|D cgc]} t#j&| d}} |d }t#j(|d}|d }t#j*|d}|d }t#j*|d}t-j.d|||||| \} } } }|dd}t#j0r{d| j3dd| j3dd| j5dd| j3dd| j3df }| j6}t#j8d||||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} wcc} w)aA queue that produces elements sorted by the first component value. Note that the PriorityQueue requires the first component of any element to be a scalar int64, in addition to the other elements declared by component_types. Therefore calls to Enqueue and EnqueueMany (resp. Dequeue and DequeueMany) on a PriorityQueue will all require (resp. output) one extra entry in their input (resp. output) lists. Args: shapes: A list of shapes (each a `tf.TensorShape` or list of `ints`). The shape of each component in a value. The length of this attr must be either 0 or the same as the length of component_types. If the length of this attr is 0, the shapes of queue elements are not constrained, and only one element may be dequeued at a time. component_types: An optional list of `tf.DTypes`. Defaults to `[]`. The type of each component in a value. capacity: An optional `int`. Defaults to `-1`. The upper bound on the number of elements in this queue. Negative numbers mean no limit. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this queue will be shared under the given name across multiple sessions. name: A name for the operation (optional). Returns: A `Tensor` of type `resource`. PriorityQueueV2rtrurorprqNrFExpected list for 'shapes' argument to 'priority_queue_v2' Op, not %r.OExpected list for 'component_types' argument to 'priority_queue_v2' Op, not %r.rmrnr)r0rr1r2rrrrrrr priority_queue_v2_eager_fallbackrryrzr{r|rIr}rfr~rr4r5rJrrrKrL)rurtrorprqr.r6r7rNrrrr8r9r:rOrPs r;priority_queue_v2r{ s>    0h..0$ #\\  11 '8/&*h Y{$gn FT5M *  *,2 3 449? ?"H  H - ?& ?O OdE] 3  *,; < ==JYY2X'',=>Y/Y H   x 4(I ;7)K!!+}=+'88&/$, '2?!QX QK' ""$.?!@(ll8$j +[ll;'ll=) +F ::L <: (' ._  & &- ##At,,  # #    -)&8;TtMM  # #    @Zr'zraw_ops.PriorityQueueV2c t|ttfstd|z|Dcgc]}t j |d}}|g}t|ttfstd|z|Dcgc]}t j |d}}|d}t j|d}|d}t j|d}|d}t j|d }g} d|d|d|d|d |f } t jd d | | || } t jrt jd | | | | \} | Scc}wcc}w)NrrurrtrmrornrprqsPriorityQueueV2rrr) ryrzr{r|rIr}rfr~rrrJrL) rurtrorprqr.r@rrrPrOrNs r;rr sw FT5M *  *,2 3 449? ?"H  H - ?& ?O OdE] 3  *,; < ==JYY2X'',=>Y/Y H   x 4(I ;7)K!!+}=+, &* KM; @&   /<#)s ?' ""$ <: (' .5 @Zr+ctjxstj}|j}|jr t d|d}t j |d}tjd|||\}}}}|S)aCloses the given queue. This operation signals that no more elements will be enqueued in the given queue. Subsequent Enqueue(Many) operations will fail. Subsequent Dequeue(Many) operations will continue to succeed if sufficient elements remain in the queue. Subsequent Dequeue(Many) operations that would block will fail immediately. Args: handle: A `Tensor` of type mutable `string`. The handle to a queue. cancel_pending_enqueues: An optional `bool`. Defaults to `False`. If true, all pending enqueue requests that are blocked on the given queue will be canceled. name: A name for the operation (optional). Returns: The created Operation. Gqueue_close op does not support eager execution. Arg 'handle' is a ref.Fr QueueCloserrrs r; queue_closer s&    0h..0$ #\\ ` aa$#$../FHab'88V.E!!QX *r=zraw_ops.QueueClosectd)Nrr?rs r;queue_close_eager_fallbackr ^__r=ctjxstj}|j}|jr t j |d||d|}|S|d}tj|d}tj d|||\}}}} |S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||S#t j$rYwxYw)aCloses the given queue. This operation signals that no more elements will be enqueued in the given queue. Subsequent Enqueue(Many) operations will fail. Subsequent Dequeue(Many) operations will continue to succeed if sufficient elements remain in the queue. Subsequent Dequeue(Many) operations that would block will fail immediately. Args: handle: A `Tensor` of type `resource`. The handle to a queue. cancel_pending_enqueues: An optional `bool`. Defaults to `False`. If true, all pending enqueue requests that are blocked on the given queue will be canceled. name: A name for the operation (optional). Returns: The created Operation. QueueCloseV2rN)rr.r@Fr)r0rr1r2rrrrrrrqueue_close_v2_eager_fallbackrrIrr4r5) r'rr.r6r7rNrr8r9r:s r;queue_close_v2r s&    0h..0$ #\\ 11 ndF,E!gn$#$../FHab'88v0G!#!QX *%  & &- ##At,,  # #    * *A  # #   s0BC)CCC!C00DDzraw_ops.QueueCloseV2c|d}tj|d}tj|tj }|g}d|f}tj dd||||}d}|S)NFrs QueueCloseV2rr)rIrrrrr,r)r'rr.r@rPrOrNs r;rrP st$#$../FHab  ! !&'*:*: ;&, %'> ?&   _a #)s ?' ' .r=cdtjxstj}|j}|jr t dt |t tfstd|z|Dcgc]}tj|d}}|d}tj|d}tjd||||\}}}} | dd} tjrHd|jdd|j!df} |j"} tj$d| | | | Scc}w) akDequeues a tuple of one or more tensors from the given queue. This operation has k outputs, where k is the number of components in the tuples stored in the given queue, and output i is the ith component of the dequeued tuple. N.B. If the queue is empty, this operation will block until an element has been dequeued (or 'timeout_ms' elapses, if specified). Args: handle: A `Tensor` of type mutable `string`. The handle to a queue. component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a tuple. timeout_ms: An optional `int`. Defaults to `-1`. If the queue is empty, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `component_types`. Iqueue_dequeue op does not support eager execution. Arg 'handle' is a ref.zKExpected list for 'component_types' argument to 'queue_dequeue' Op, not %r.rtNrmr QueueDequeuer'rtrr.r0rr1r2r3ryrzr{r|rIrfr~r4r5rJrrrKrL) r'rtrr.r6r7rr8r9r:rNrOrPs r; queue_dequeuer] s2.    0h..0$ #\\ b cc OdE] 3  &(7 8 99JYY2X'',=>Y/YJ  \:*'88v#-D:!QX QK' ""$.?!@C--l;=F::L  fg7 .Zs2D-zraw_ops.QueueDequeuectd)Nrr?)r'rtrr.r@s r;queue_dequeue_eager_fallbackr rr=ncftjxstj}|j}|jr t dt |t tfstd|z|Dcgc]}tj|d}}|d}tj|d}tjd|||||\}}} } | dd} tjrHd| jdd| j!df} | j"} tj$d| | | | Scc}w) a Dequeues `n` tuples of one or more tensors from the given queue. If the queue is closed and there are fewer than `n` elements, then an OutOfRange error is returned. This operation concatenates queue-element component tensors along the 0th dimension to make a single component tensor. All of the components in the dequeued tuple will have size `n` in the 0th dimension. This operation has `k` outputs, where `k` is the number of components in the tuples stored in the given queue, and output `i` is the ith component of the dequeued tuple. N.B. If the queue is empty, this operation will block until `n` elements have been dequeued (or 'timeout_ms' elapses, if specified). Args: handle: A `Tensor` of type mutable `string`. The handle to a queue. n: A `Tensor` of type `int32`. The number of tuples to dequeue. component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a tuple. timeout_ms: An optional `int`. Defaults to `-1`. If the queue has fewer than n elements, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `component_types`. Nqueue_dequeue_many op does not support eager execution. Arg 'handle' is a ref.zPExpected list for 'component_types' argument to 'queue_dequeue_many' Op, not %r.rtNrmrQueueDequeueManyr'rrtrr.rr'rrtrr.r6r7rr8r9r:rNrOrPs r;queue_dequeue_manyr s5>    0h..0$ #\\ g hh OdE] 3  +-< = >>JYY2X'',=>Y/YJ  \:*'886Q,;'1>!QX QK' ""$.?!@C--l;=F::L L&'; .Z2D.zraw_ops.QueueDequeueManyctd)Nrr?r'rrtrr.r@s r;!queue_dequeue_many_eager_fallbackr  rr=c tjxstj}|j}|jr t j |d|||d|d| }|St|ttfst!d|z|D cgc]} t#j$| d}} |d}t#j&|d}t)j*d|||||\} } } } | dd}|s| St#j,rHd| j/dd| j1df} | j2}t#j4d|| ||S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||||S#t j$rYZwxYwcc} w) aDequeues `n` tuples of one or more tensors from the given queue. If the queue is closed and there are fewer than `n` elements, then an OutOfRange error is returned. This operation concatenates queue-element component tensors along the 0th dimension to make a single component tensor. All of the components in the dequeued tuple will have size `n` in the 0th dimension. This operation has `k` outputs, where `k` is the number of components in the tuples stored in the given queue, and output `i` is the ith component of the dequeued tuple. N.B. If the queue is empty, this operation will block until `n` elements have been dequeued (or 'timeout_ms' elapses, if specified). Args: handle: A `Tensor` of type `resource`. The handle to a queue. n: A `Tensor` of type `int32`. The number of tuples to dequeue. component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a tuple. timeout_ms: An optional `int`. Defaults to `-1`. If the queue has fewer than n elements, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `component_types`. QueueDequeueManyV2rtrNrtrr.r@SExpected list for 'component_types' argument to 'queue_dequeue_many_v2' Op, not %r.rmr)r0rr1r2rrrrrrr$queue_dequeue_many_v2_eager_fallbackrryrzr{r|rIrfr~r4r5rJrrrKrLr'rrtrr.r6r7rNrrr8r9r:rOrPs r;queue_dequeue_many_v2r s>    0h..0$ #\\ 11 "D&!5Fz3gn OdE] 3  .0? @ AAJYY2X'',=>Y/YJ  \:*'88Vq.=)3$@!QX QK'  J ""$.?!@C--l;=F::L lFG= .A  & &- ##At,,  # #    1 !_  # #   Z6EF=FE55F FF##F:9F:zraw_ops.QueueDequeueManyV2ct|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}tj|tj}tj|tj}||g}d|d|f}t jdt|||||} t jrt jd||| | Scc}w)NrrtrmrsQueueDequeueManyV2rr ryrzr{r|rIrfr~rrrr,rrrrJrL r'rrtrr.r@rrPrOrNs r;rrs OdE] 3  .0? @ AAJYY2X'',=>Y/YJ  \:*  ! !&'*:*: ;& Q .!!,  j I&   2C4H$0C"& ('""$ lFG= .ZDcftjxstj}|j}|jr t dt |t tfstd|z|Dcgc]}tj|d}}|d}tj|d}tjd|||||\}}} } | dd} tjrHd| jdd| j!df} | j"} tj$d| | | | Scc}w) a6Dequeues `n` tuples of one or more tensors from the given queue. This operation is not supported by all queues. If a queue does not support DequeueUpTo, then an Unimplemented error is returned. If the queue is closed and there are more than 0 but less than `n` elements remaining, then instead of returning an OutOfRange error like QueueDequeueMany, less than `n` elements are returned immediately. If the queue is closed and there are 0 elements left in the queue, then an OutOfRange error is returned just like in QueueDequeueMany. Otherwise the behavior is identical to QueueDequeueMany: This operation concatenates queue-element component tensors along the 0th dimension to make a single component tensor. All of the components in the dequeued tuple will have size `n` in the 0th dimension. This operation has k outputs, where `k` is the number of components in the tuples stored in the given queue, and output `i` is the ith component of the dequeued tuple. Args: handle: A `Tensor` of type mutable `string`. The handle to a queue. n: A `Tensor` of type `int32`. The number of tuples to dequeue. component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a tuple. timeout_ms: An optional `int`. Defaults to `-1`. If the queue has fewer than n elements, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `component_types`. Oqueue_dequeue_up_to op does not support eager execution. Arg 'handle' is a ref.zQExpected list for 'component_types' argument to 'queue_dequeue_up_to' Op, not %r.rtNrmrQueueDequeueUpTorrrs r;queue_dequeue_up_tor4s6F    0h..0$ #\\ h ii OdE] 3  ,.= > ??JYY2X'',=>Y/YJ  \:*'886Q,;'1>!QX QK' ""$.?!@C--l;=F::L L&'; .Zrzraw_ops.QueueDequeueUpToctd)Nrr?r s r;"queue_dequeue_up_to_eager_fallbackrtrr=c tjxstj}|j}|jr t j |d|||d|d| }|St|ttfst!d|z|D cgc]} t#j$| d}} |d}t#j&|d}t)j*d|||||\} } } } | dd}|s| St#j,rHd| j/dd| j1df} | j2}t#j4d|| ||S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||||S#t j$rYZwxYwcc} w) a0Dequeues `n` tuples of one or more tensors from the given queue. This operation is not supported by all queues. If a queue does not support DequeueUpTo, then an Unimplemented error is returned. If the queue is closed and there are more than 0 but less than `n` elements remaining, then instead of returning an OutOfRange error like QueueDequeueMany, less than `n` elements are returned immediately. If the queue is closed and there are 0 elements left in the queue, then an OutOfRange error is returned just like in QueueDequeueMany. Otherwise the behavior is identical to QueueDequeueMany: This operation concatenates queue-element component tensors along the 0th dimension to make a single component tensor. All of the components in the dequeued tuple will have size n in the 0th dimension. This operation has `k` outputs, where `k` is the number of components in the tuples stored in the given queue, and output `i` is the ith component of the dequeued tuple. Args: handle: A `Tensor` of type `resource`. The handle to a queue. n: A `Tensor` of type `int32`. The number of tuples to dequeue. component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a tuple. timeout_ms: An optional `int`. Defaults to `-1`. If the queue has fewer than n elements, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `component_types`. QueueDequeueUpToV2rtrNrTExpected list for 'component_types' argument to 'queue_dequeue_up_to_v2' Op, not %r.rmr)r0rr1r2rrrrrrr%queue_dequeue_up_to_v2_eager_fallbackrryrzr{r|rIrfr~r4r5rJrrrKrLrs r;queue_dequeue_up_to_v2r"wsF    0h..0$ #\\ 11 "D&!5Fz3gn OdE] 3  /1@ A BBJYY2X'',=>Y/YJ  \:*'88Vq.=)3$@!QX QK'  J ""$.?!@C--l;=F::L lFG= .A  & &- ##At,,  # #    2 !_  # #   Zrzraw_ops.QueueDequeueUpToV2ct|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}tj|tj}tj|tj}||g}d|d|f}t jdt|||||} t jrt jd||| | Scc}w)Nr rtrmrsQueueDequeueUpToV2rrrrs r;r!r!s OdE] 3  /1@ A BBJYY2X'',=>Y/YJ  \:*  ! !&'*:*: ;& Q .!!,  j I&   2C4H$0C"& ('""$ lFG= .Zrc ~tjxstj}|j}|jr t j |d||d|d|}|St|ttfst!d|z|Dcgc]}t#j$|d}}|d}t#j&|d}t)j*d||||\} } } } | dd}|s| St#j,rHd| j/dd| j1df} | j2} t#j4d| | ||S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||||S#t j$rYXwxYwcc}w) aeDequeues a tuple of one or more tensors from the given queue. This operation has k outputs, where k is the number of components in the tuples stored in the given queue, and output i is the ith component of the dequeued tuple. N.B. If the queue is empty, this operation will block until an element has been dequeued (or 'timeout_ms' elapses, if specified). Args: handle: A `Tensor` of type `resource`. The handle to a queue. component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a tuple. timeout_ms: An optional `int`. Defaults to `-1`. If the queue is empty, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `component_types`. QueueDequeueV2rtrNrNExpected list for 'component_types' argument to 'queue_dequeue_v2' Op, not %r.rmr)r0rr1r2rrrrrrrqueue_dequeue_v2_eager_fallbackrryrzr{r|rIrfr~r4r5rJrrrKrL)r'rtrr.r6r7rNrrr8r9r:rOrPs r;queue_dequeue_v2r(s.    0h..0$ #\\ 11 f.?z3gn OdE] 3  )+: ; <<JYY2X'',=>Y/YJ  \:*'88%/d<!QX QK'  J ""$.?!@C--l;=F::L ,9 .?  & &- ##At,,  # #    , /j  # #   Zs6EF:F E33F  F F F76F7zraw_ops.QueueDequeueV2ct|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}tj|tj}|g}d|d|f}t jdt|||||}t jrt jd||||Scc}w)Nr&rtrmrsQueueDequeueV2rr%)ryrzr{r|rIrfr~rrrr,rrrJrL) r'rtrr.r@rrPrOrNs r;r'r' s OdE] 3  )+: ; <<JYY2X'',=>Y/YJ  \:*  ! !&'*:*: ;&,  j I&   .O0D$0C"& ('""$ ,9 .ZsC!c tjxstj}|j}|jr t d|d}t j |d}tjd||||\}}}}|S)aEnqueues a tuple of one or more tensors in the given queue. The components input has k elements, which correspond to the components of tuples stored in the given queue. N.B. If the queue is full, this operation will block until the given element has been enqueued (or 'timeout_ms' elapses, if specified). Args: handle: A `Tensor` of type mutable `string`. The handle to a queue. components: A list of `Tensor` objects. One or more tensors from which the enqueued tensors should be taken. timeout_ms: An optional `int`. Defaults to `-1`. If the queue is full, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: The created Operation. Iqueue_enqueue op does not support eager execution. Arg 'handle' is a ref.rmr QueueEnqueuer' componentsrr.r r'r.rr.r6r7r8r9r:s r; queue_enqueuer05s,    0h..0$ #\\ b ccJ  \:*'88v*#-D:!QX *r=zraw_ops.QueueEnqueuectd)Nr+r?r'r.rr.r@s r;queue_enqueue_eager_fallbackr3Zrr=c tjxstj}|j}|jr t d|d}t j |d}tjd||||\}}}}|S)aEnqueues zero or more tuples of one or more tensors in the given queue. This operation slices each component tensor along the 0th dimension to make multiple queue elements. All of the tuple components must have the same size in the 0th dimension. The components input has k elements, which correspond to the components of tuples stored in the given queue. N.B. If the queue is full, this operation will block until the given elements have been enqueued (or 'timeout_ms' elapses, if specified). Args: handle: A `Tensor` of type mutable `string`. The handle to a queue. components: A list of `Tensor` objects. One or more tensors from which the enqueued tensors should be taken. timeout_ms: An optional `int`. Defaults to `-1`. If the queue is too full, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: The created Operation. Nqueue_enqueue_many op does not support eager execution. Arg 'handle' is a ref.rmrQueueEnqueueManyr-rr/s r;queue_enqueue_manyr7]s6    0h..0$ #\\ g hhJ  \:*'886j'1>!QX *r=zraw_ops.QueueEnqueueManyctd)Nr5r?r2s r;!queue_enqueue_many_eager_fallbackr9rr=c tjxstj}|j}|jr t j |d|||d|}|S|d}tj|d}tj d||||\}}} } | S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||||S#t j$rYwxYw)aEnqueues zero or more tuples of one or more tensors in the given queue. This operation slices each component tensor along the 0th dimension to make multiple queue elements. All of the tuple components must have the same size in the 0th dimension. The components input has k elements, which correspond to the components of tuples stored in the given queue. N.B. If the queue is full, this operation will block until the given elements have been enqueued (or 'timeout_ms' elapses, if specified). Args: handle: A `Tensor` of type `resource`. The handle to a queue. components: A list of `Tensor` objects. One or more tensors from which the enqueued tensors should be taken. timeout_ms: An optional `int`. Defaults to `-1`. If the queue is too full, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: The created Operation. QueueEnqueueManyV2rNrr.r@rmr-)r0rr1r2rrrrrrr$queue_enqueue_many_v2_eager_fallbackrrIr~r4r5 r'r.rr.r6r7rNrr8r9r:s r;queue_enqueue_many_v2r?s6    0h..0$ #\\ 11 "D&*lgnJ  \:*'88V )3$@!QX *!  & &- ##At,,  # #    1 *$DJJ  # #   0BC+CCC#C33D D zraw_ops.QueueEnqueueManyV2c|d}tj|d}tj||\}}tj|t j }|gt|z}d|d|f}tjdd||||}d}|S)Nrmr TcomponentssQueueEnqueueManyV2rr rIr~rrrrr,rzr r'r.rr.r@_attr_TcomponentsrPrOrNs r;r=r=sJ  \:*"*"I"I*VY"ZZ  ! !&'*:*: ;&D,,, ,lJ G&   2Al#)s ?' ' .r=c tjxstj}|j}|jr t j |d|||d|}|S|d}tj|d}tj d||||\}}} } | S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||||S#t j$rYwxYw)aEnqueues a tuple of one or more tensors in the given queue. The components input has k elements, which correspond to the components of tuples stored in the given queue. N.B. If the queue is full, this operation will block until the given element has been enqueued (or 'timeout_ms' elapses, if specified). Args: handle: A `Tensor` of type `resource`. The handle to a queue. components: A list of `Tensor` objects. One or more tensors from which the enqueued tensors should be taken. timeout_ms: An optional `int`. Defaults to `-1`. If the queue is full, this operation will block for up to timeout_ms milliseconds. Note: This option is not supported yet. name: A name for the operation (optional). Returns: The created Operation. QueueEnqueueV2rNr<rmr-)r0rr1r2rrrrrrrqueue_enqueue_v2_eager_fallbackrrIr~r4r5r>s r;queue_enqueue_v2rIs,    0h..0$ #\\ 11 fj,gnJ  \:*'88J%/d<!QX *!  & &- ##At,,  # #    , *$DJJ  # #   r@zraw_ops.QueueEnqueueV2c|d}tj|d}tj||\}}tj|t j }|gt|z}d|d|f}tjdd||||}d}|S)NrmrrBsQueueEnqueueV2rrrCrDs r;rHrHsJ  \:*"*"I"I*VY"ZZ  ! !&'*:*: ;&D,,, ,lJ G&   .,#)s ?' ' .r=cXtjxstj}|j}|jr t dt j d||\}}}}|dd}tjr&d}|j} tjd| |||\}|S)a&Returns true if queue is closed. This operation returns true if the queue is closed and false if the queue is open. Args: handle: A `Tensor` of type mutable `string`. The handle to a queue. name: A name for the operation (optional). Returns: A `Tensor` of type `bool`. Kqueue_is_closed op does not support eager execution. Arg 'handle' is a ref. QueueIsClosedrFNrGrHrMs r;queue_is_closedrNs    0h..0$ #\\ d ee'88T3!QX QK' ""$ F::L vw8 (' .r=zraw_ops.QueueIsClosedctd)NrLr?rSs r;queue_is_closed_eager_fallbackrP/sbccr=c\tjxstj}|j}|jr t j |d||}|Stjd||\}}}}|dd}tj r&d} |j"} tj$d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||S#t j$rYwxYw)a Returns true if queue is closed. This operation returns true if the queue is closed and false if the queue is open. Args: handle: A `Tensor` of type `resource`. The handle to a queue. name: A name for the operation (optional). Returns: A `Tensor` of type `bool`. QueueIsClosedV2NrrFrG)r0rr1r2rrrrrrr!queue_is_closed_v2_eager_fallbackrr4r5rIrJrKrL r'r.r6r7rNrr8r9r:rOrPs r;queue_is_closed_v2rU2s(    0h..0$ #\\ 11 v/g n(88&t5!QX QK' ""$ F::L <: (' .'  & &- ##At,,  # #    . t''  # #   0B<<DC**DD DD+*D+zraw_ops.QueueIsClosedV2ctj|tj}|g}d}t j dd||||}t j rt jd||||\}|S)NsQueueIsClosedV2rrrRrrrr,rIrrJrLrs r;rSrS^ss  ! !&'*:*: ;&, &   /<#)s ?' ""$ <: (' .r=cXtjxstj}|j}|jr t dt j d||\}}}}|dd}tjr&d}|j} tjd| |||\}|S)zComputes the number of elements in the given queue. Args: handle: A `Tensor` of type mutable `string`. The handle to a queue. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. Fqueue_size op does not support eager execution. Arg 'handle' is a ref. QueueSizerFNrGrHrMs r; queue_sizer\ks    0h..0$ #\\ _ ``'88F/!QX QK' ""$ F::L \674 (' .r=zraw_ops.QueueSizectd)NrZr?rSs r;queue_size_eager_fallbackr^rr=c\tjxstj}|j}|jr t j |d||}|Stjd||\}}}}|dd}tj r&d} |j"} tj$d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||S#t j$rYwxYw)zComputes the number of elements in the given queue. Args: handle: A `Tensor` of type `resource`. The handle to a queue. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. QueueSizeV2NrrFrG)r0rr1r2rrrrrrrqueue_size_v2_eager_fallbackrr4r5rIrJrKrLrTs r; queue_size_v2rbs'    0h..0$ #\\ 11 mT6+g n(88f41!QX QK' ""$ F::L |VW6 (' .'  & &- ##At,,  # #    ) t''  # #   rVzraw_ops.QueueSizeV2ctj|tj}|g}d}t j dd||||}t j rt jd||||\}|S)Ns QueueSizeV2rrr`rXrs r;rarasr  ! !&'*:*: ;&, &   ^Q|#)s ?' ""$ |VW6 (' .r=min_after_dequeueseedseed2c tjxstj} | j} | jr t dt |t tfstd|z|D cgc]} tj| d}} |g}t |t tfstd|z|D cgc]} tj| d}} |d}tj|d}|d }tj|d }|d }tj|d }|d }tj|d }|d }tj|d}|d }tj|d}tjd||||||||| \} } }}|dd}tj rd|j#dd|j#dd|j%dd |j%d d |j%d d |j%d d|j#dd|j#df}|j&}tj(d||||\}|Scc} wcc} w)apA queue that randomizes the order of elements. Args: component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a value. shapes: An optional list of shapes (each a `tf.TensorShape` or list of `ints`). Defaults to `[]`. The shape of each component in a value. The length of this attr must be either 0 or the same as the length of component_types. If the length of this attr is 0, the shapes of queue elements are not constrained, and only one element may be dequeued at a time. capacity: An optional `int`. Defaults to `-1`. The upper bound on the number of elements in this queue. Negative numbers mean no limit. min_after_dequeue: An optional `int`. Defaults to `0`. Dequeue will block unless there would be this many elements after the dequeue or the queue is closed. This ensures a minimum level of mixing of elements. seed: An optional `int`. Defaults to `0`. If either seed or seed2 is set to be non-zero, the random number generator is seeded by the given seed. Otherwise, a random seed is used. seed2: An optional `int`. Defaults to `0`. A second seed to avoid seed collision. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this queue will be shared under the given name across multiple sessions. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Prandom_shuffle_queue op does not support eager execution. Arg 'handle' is a ref.zRExpected list for 'component_types' argument to 'random_shuffle_queue' Op, not %r.rtNzIExpected list for 'shapes' argument to 'random_shuffle_queue' Op, not %r.rurmrorrdrerfrnrprqRandomShuffleQueue rtrurordrerfrprqr.rx)rtrurordrerfrprqr.r6r7rrr8r9r:rNrOrPs r;random_shuffle_queuerksD    0h..0$ #\\ i jj OdE] 3  -/> ? @@JYY2X'',=>Y/Y ^ F FT5M *  -/5 6 779? ?"H  H - ?& ? H   x 4(''(9;NO \ D   4 ($ ] E   E7 +%I ;7)K!!+}=+'88of'/0A$)Y*5D B!QX QK' ""$.?!@(ll8$j +-@ 34f'#2C2CG2L3<< 4mll=) +F::L lFG= (' .[Z @s 2I)=I.zraw_ops.RandomShuffleQueuec td)Nrhr? rtrurordrerfrprqr.r@s r;#random_shuffle_queue_eager_fallbackrnghhr=c tjxstj} | j} | jr* t j | d|d|d|d|d|d|d|d|d |} | St|ttfst!d |z|D cgc]} t#j$| d}} |g}t|ttfst!d |z|Dcgc]}t#j&|d}}|d}t#j(|d}|d}t#j(|d}|d}t#j(|d}|d}t#j(|d}|d}t#j*|d}|d}t#j*|d }t-j.d||||||||| \}}}}|d d } t#j0rd|j3dd|j3dd|j5dd|j5dd|j5dd|j5dd|j3dd |j3d f}|j6}t#j8d||| | \} | S#t j$r } tj| |Yd } ~ nd } ~ wt j$rYnwxYw t|||||||||| S#t j$rYwxYwcc} wcc}w)ajA queue that randomizes the order of elements. Args: component_types: A list of `tf.DTypes` that has length `>= 1`. The type of each component in a value. shapes: An optional list of shapes (each a `tf.TensorShape` or list of `ints`). Defaults to `[]`. The shape of each component in a value. The length of this attr must be either 0 or the same as the length of component_types. If the length of this attr is 0, the shapes of queue elements are not constrained, and only one element may be dequeued at a time. capacity: An optional `int`. Defaults to `-1`. The upper bound on the number of elements in this queue. Negative numbers mean no limit. min_after_dequeue: An optional `int`. Defaults to `0`. Dequeue will block unless there would be this many elements after the dequeue or the queue is closed. This ensures a minimum level of mixing of elements. seed: An optional `int`. Defaults to `0`. If either seed or seed2 is set to be non-zero, the random number generator is seeded by the given seed. Otherwise, a random seed is used. seed2: An optional `int`. Defaults to `0`. A second seed to avoid seed collision. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this queue will be shared under the given name across multiple sessions. name: A name for the operation (optional). Returns: A `Tensor` of type `resource`. RandomShuffleQueueV2rtrurordrerfrprqNrmUExpected list for 'component_types' argument to 'random_shuffle_queue_v2' Op, not %r.LExpected list for 'shapes' argument to 'random_shuffle_queue_v2' Op, not %r.rmrrnrj)r0rr1r2rrrrrrr&random_shuffle_queue_v2_eager_fallbackrryrzr{r|rIrfr}r~rr4r5rJrrrKrL)rtrurordrerfrprqr.r6r7rNrrrr8r9r:rOrPs r;random_shuffle_queue_v2ru!sUD    0h..0$ #\\  11 $d,=6:x.guY { Y/Y ^ F FT5M *  028 9 ::9? ?"H  H - ?& ? H   x 4(''(9;NO \ D   4 ($ ] E   E7 +%I ;7)K!!+}=+'88'-2C%)),7d D!QX QK' ""$.?!@(ll8$j +-@ 34f'#2C2CG2L3<< 4mll=) +F::L  fg? (' .{  & &- ##At,,  # #    3)&8-D;TtMM  # #   Z @s<(JLLKJ66KKK((K?>K?zraw_ops.RandomShuffleQueueV2c 4t|ttfstd|z|D cgc]} t j | d}} |g}t|ttfstd|z|D cgc]} t j | d}} |d}t j|d}|d}t j|d}|d}t j|d }|d}t j|d }|d }t j|d }|d }t j|d }g} d|d|d|d|d |d |d |d |f} t jdd| | | |}t jrt jd| | ||\}|Scc} wcc} w)NrrrtrsrurmrorrdrerfrnrprqsRandomShuffleQueueV2rrrqr))rtrurordrerfrprqr.r@rrrPrOrNs r;rtrts OdE] 3  02A B CCJYY2X'',=>Y/Y ^ F FT5M *  028 9 ::9? ?"H  H - ?& ? H   x 4(''(9;NO \ D   4 ($ ] E   E7 +%I ;7)K!!+}=+, &* !2FD'i  =&   4a #)s ?' ""$  fg? (' .IZ @s F4F file_patternfile_random_seedfile_shuffle_shift_ratiofile_buffer_sizefile_parallelism batch_sizecompression_typectjxstj}|j} | jr( t j |d|d|d|d|d|d|d|d|} | Stj|d}|d }tj|d}|d }tj |d}|d }tj|d}|d}tj|d}|d}tj|d}|d}tj|d}t#j$d|||||||| \} } } }|d d } tj&rd| j)dd| j+dd| j)dd| j+dd| j+dd| j+dd| j)df}| j,}tj.d||| | \} | S#t j$r } tj| |Yd } ~ nd } ~ wt j$rYnwxYw t||||||||| S#t j$rYwxYw)aEmits randomized records. Args: file_pattern: A `string`. Glob pattern for the data files. file_random_seed: An optional `int`. Defaults to `301`. Random seeds used to produce randomized records. file_shuffle_shift_ratio: An optional `float`. Defaults to `0`. Shifts the list of files after the list is randomly shuffled. file_buffer_size: An optional `int`. Defaults to `10000`. The randomization shuffling buffer. file_parallelism: An optional `int`. Defaults to `16`. How many sstables are opened and concurrently iterated over. batch_size: An optional `int`. Defaults to `32`. The batch size. compression_type: An optional `string`. Defaults to `""`. The type of compression for the file. Currently ZLIB and GZIP are supported. Defaults to none. name: A name for the operation (optional). Returns: A `Tensor` of type `string`. RecordInputrwrxryrzr{r|r}N) rwrxryrzr{r|r}r.r@-r' rn)rwrxryrzr{r|r}r.)r0rr1r2rrrrrrrrecord_input_eager_fallbackrrIrr~ make_floatr4r5rJrrrKrL)rwrxryrzr{r|r}r.r6r7rNrr8r9r:rOrPs r; record_inputrs.    0h..0$ #\\  11 mT><,.H "46F,lJ, .g n""<@,&&'79KL% %001IKef&&'79KL&&'79KLJ  \:*&&'79KL'88L(80H(8(8",(8tE!QX QK' ""$cll>: #"3"34F"G(ll568J 235G 23\ -/All-.0F::L |VW6 (' .o  & &- ##At,,  # #    (#6F#;++ +$D BB  # #   s0&G88H? H&&H?>H?II.-I.zraw_ops.RecordInputc $tj|d}|d}tj|d}|d}tj|d}|d}tj|d}|d}tj|d }|d }tj|d }|d }tj|d }g} d|d|d|d|d |d |d |f} tjdd| | ||} tj rtj d| | | | \} | S)Nrwrrxrryrrzrr{rr|rnr}s RecordInputrrr)rIrr~rrrJrL) rwrxryrzr{r|r}r.r@rPrOrNs r;rrsY""<@,&&'79KL% %001IKef&&'79KL&&'79KLJ  \:*&&'79KL, L*<.0H&(: ,e$DBB  # #   0C((D/;DD/.D/3EEEz'raw_ops.ResourceAccumulatorTakeGradientchtj|d}tj|tj }tj|tj }||g}d|f}tjdd||||}tjrtjd||||\}|S)Nr`sResourceAccumulatorTakeGradientrrr) rIrfrrrr,rrrJrL)r'r_r`r.r@rPrOrNs r;rrs   UG ,%  ! !&'*:*: ;&'' gmmD,,', U &   ?$0C"& ('""$ )<J (' .r='TV_ResourceConditionalAccumulator_dtypectjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|Stj|d}tj|d}|d }tj |d}|d }tj |d}|d }tj |d}t#j$d|||||| \} } } } | dd}tj&r{d| j)dd| j+dd| j+dd| j+dd| j+df } | j,}tj.d|| ||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYw) aA conditional accumulator for aggregating gradients. The accumulator accepts gradients marked with local_step greater or equal to the most recent global_step known to the accumulator. The average can be extracted from the accumulator, provided sufficient gradients have been accumulated. Extracting the average automatically resets the aggregate to 0, and increments the global_step recorded by the accumulator. This is a resource version of ConditionalAccumulator that will work in TF2.0 with tf.cond version 2. Args: dtype: A `tf.DType` from: `tf.float32, tf.float64, tf.int32, tf.uint8, tf.int16, tf.int8, tf.complex64, tf.int64, tf.qint8, tf.quint8, tf.qint32, tf.bfloat16, tf.qint16, tf.quint16, tf.uint16, tf.complex128, tf.half, tf.uint32, tf.uint64`. The type of the value being accumulated. shape: A `tf.TensorShape` or list of `ints`. The shape of the values, can be [], in which case shape is unknown. container: An optional `string`. Defaults to `""`. If non-empty, this accumulator is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this accumulator will be shared under the given name across multiple sessions. reduction_type: An optional `string` from: `"MEAN", "SUM"`. Defaults to `"MEAN"`. name: A name for the operation (optional). Returns: A `Tensor` of type `resource`. ResourceConditionalAccumulatorr`rrprqrNrrnrr)r0rr1r2rrrrrrr/resource_conditional_accumulator_eager_fallbackrrIrfr}rr4r5rJrgrrKrL)r`rrprqrr.r6r7rNrr8r9r:rOrPs r; resource_conditional_accumulatorr,s#:    0h..0$ #\\  11 .gug {I}k.*gn   UG ,%   eW -%I ;7)K!!+}=+N$$^5EF.'88(U4=6A9G/3 5!QX QK' ""$s))'2Gll7#[#,,{2KS\\-8:Jll+,.F::L (,I (' .O  & &- ##At,,  # #    <Ui!.t  # #   s0"FG+GGG#G55H  H z&raw_ops.ResourceConditionalAccumulatorc tj|d}tj|d}|d}tj|d}|d}tj|d}|d}tj|d}g}d|d|d|d|d|f }tjdd |||| } tj rtj d ||| | \} | S) Nr`rrnrprqrrsResourceConditionalAccumulatorrrr)rIrfr}rrrJrL) r`rrprqrr.r@rPrOrNs r;rr~s   UG ,%   eW -%I ;7)K!!+}=+N$$^5EF., UGUK. @&   >$0C"& ('""$ (,I (' .r='TV_SparseAccumulatorApplyGradient_dtypegradient_indicesgradient_valuesgradient_shapehas_known_shapec tjxstj}|j}|jr t dt j |d}tjd|||||||\} } } } | S)aApplies a sparse gradient to a given accumulator. Does not add if local_step is smaller than the accumulator's global_step. Args: handle: A `Tensor` of type mutable `string`. The handle to a accumulator. local_step: A `Tensor` of type `int64`. The local_step value at which the sparse gradient was computed. gradient_indices: A `Tensor` of type `int64`. Indices of the sparse gradient to be accumulated. Must be a vector. gradient_values: A `Tensor`. Must be one of the following types: `float32`, `float64`, `int32`, `uint8`, `int16`, `int8`, `complex64`, `int64`, `qint8`, `quint8`, `qint32`, `bfloat16`, `qint16`, `quint16`, `uint16`, `complex128`, `half`, `uint32`, `uint64`. Values are the non-zero slices of the gradient, and must have the same first dimension as indices, i.e., the nnz represented by indices and values must be consistent. gradient_shape: A `Tensor` of type `int64`. Shape of the sparse gradient to be accumulated. has_known_shape: A `bool`. Boolean indicating whether gradient_shape is unknown, in which case the input is ignored during validation. name: A name for the operation (optional). Returns: The created Operation. ]sparse_accumulator_apply_gradient op does not support eager execution. Arg 'handle' is a ref.rSparseAccumulatorApplyGradient)r'r(rrrrr.r) r'r(rrrrr.r6r7r8r9r:s r;!sparse_accumulator_apply_gradientrs6    0h..0$ #\\ v ww&&8IJ/'88(5?;K:I9G:I/35!QX *r=z&raw_ops.SparseAccumulatorApplyGradientctd)Nrr?)r'r(rrrrr.r@s r;0sparse_accumulator_apply_gradient_eager_fallbackrstuur=SparseAccumulatorTakeGradient)rrr&TV_SparseAccumulatorTakeGradient_dtypectjxstj}|j}|jr t dt j |d}tjd||||\}}}}|dd} t jr7d|jdf} |j} t jd| | | tj| } | S)aExtracts the average sparse gradient in a SparseConditionalAccumulator. The op will blocks until sufficient (i.e., more than num_required) gradients have been accumulated. If the accumulator has already aggregated more than num_required gradients, it will return its average of the accumulated gradients. Also automatically increments the recorded global_step in the accumulator by 1, and resets the aggregate to 0. Args: handle: A `Tensor` of type mutable `string`. The handle to a SparseConditionalAccumulator. num_required: A `Tensor` of type `int32`. Number of gradients required before we return an aggregate. dtype: A `tf.DType` from: `tf.float32, tf.float64, tf.int32, tf.uint8, tf.int16, tf.int8, tf.complex64, tf.int64, tf.qint8, tf.quint8, tf.qint32, tf.bfloat16, tf.qint16, tf.quint16, tf.uint16, tf.complex128, tf.half, tf.uint32, tf.uint64`. The data type of accumulated gradients. Needs to correspond to the type of the accumulator. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (indices, values, shape). indices: A `Tensor` of type `int64`. values: A `Tensor` of type `dtype`. shape: A `Tensor` of type `int64`. \sparse_accumulator_take_gradient op does not support eager execution. Arg 'handle' is a ref.r`rrdN)r0rr1r2r3rIrfr4r5rJrgrKrL$_SparseAccumulatorTakeGradientOutputrrhs r; sparse_accumulator_take_gradientrs6    0h..0$ #\\ u vv   UG ,%'88'6B/44A!QX QK' ""$s))'2 3F::L 'vwH 0 6 6w ?' .r=z%raw_ops.SparseAccumulatorTakeGradientctd)Nrr?rks r;/sparse_accumulator_take_gradient_eager_fallbackrssttr=%TV_SparseConditionalAccumulator_dtypec tjxstj}|j}|jr t dt j |d}t j|d}|d}t j|d}|d}t j|d}|d}t j|d }tjd |||||| \}}} } | dd} t jr{d| jdd| jdd| jdd| jdd | jd f } | j} t jd | | | | \} | S) a0A conditional accumulator for aggregating sparse gradients. The accumulator accepts gradients marked with local_step greater or equal to the most recent global_step known to the accumulator. The average can be extracted from the accumulator, provided sufficient gradients have been accumulated. Extracting the average automatically resets the aggregate to 0, and increments the global_step recorded by the accumulator. Args: dtype: A `tf.DType` from: `tf.float32, tf.float64, tf.int32, tf.uint8, tf.int16, tf.int8, tf.complex64, tf.int64, tf.qint8, tf.quint8, tf.qint32, tf.bfloat16, tf.qint16, tf.quint16, tf.uint16, tf.complex128, tf.half, tf.uint32, tf.uint64`. The type of the value being accumulated. shape: A `tf.TensorShape` or list of `ints`. The shape of the values. container: An optional `string`. Defaults to `""`. If non-empty, this accumulator is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. If non-empty, this accumulator will be shared under the given name across multiple sessions. reduction_type: An optional `string` from: `"MEAN", "SUM"`. Defaults to `"MEAN"`. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Zsparse_conditional_accumulator op does not support eager execution. Arg 'handle' is a ref.r`rNrnrprqrrSparseConditionalAccumulatorrrrs r;sparse_conditional_accumulatorrsy4    0h..0$ #\\ s tt   UG ,%   eW -%I ;7)K!!+}=+N$$^5EF.'88&e52;4?7E-1 3!QX QK' ""$s))'2Gll7#[#,,{2KS\\-8:Jll+,.F::L & fgG (' .r=z$raw_ops.SparseConditionalAccumulatorctd)Nrr?rs r;-sparse_conditional_accumulator_eager_fallbackrDsqrrr=) TV_Stack_elem_typerrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&r elem_type stack_namectjxstj}|j}|jr t dt j |d}|d}t j|d}tjd|||\}}}}|dd}t jrHd|jdd|jdf} |j} t jd| | ||\}|S)zDeprecated, use StackV2. Args: elem_type: A `tf.DType`. stack_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Astack op does not support eager execution. Arg 'handle' is a ref.rNrnrStack)rrr.)r0rr1r2r3rIrfrr4r5rJrgrrKrL) rrr.r6r7r8r9r:rNrOrPs r;_stackrIs    0h..0$ #\\ Z [[  K8)J  \:*'889$H!QX QK' ""$3--k:Lll<(*F::L vw0 (' .r=z raw_ops.Stackctd)Nrr?rrr.r@s r;_stack_eager_fallbackrlsXYYr=ctjxstj}|j}|jr t dt j d||\}}}}|S)zDeprecated, use StackCloseV2. Args: handle: A `Tensor` of type mutable `string`. name: A name for the operation (optional). Returns: The created Operation. Gstack_close op does not support eager execution. Arg 'handle' is a ref. StackCloserFr/r'r.r6r7r8r9r:s r; stack_closeros`    0h..0$ #\\ ` aa'88V$0!QX *r=zraw_ops.StackClosectd)Nrr?rSs r;stack_close_eager_fallbackrrr=ctjxstj}|j}|jr t j |d||}|Stjd||\}}}}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||S#t j$rYwxYw)zDelete the stack from its resource container. Args: handle: A `Tensor` of type `resource`. The handle to a stack. name: A name for the operation (optional). Returns: The created Operation. StackCloseV2NrrF)r0rr1r2rrrrrrrstack_close_v2_eager_fallbackrr4r5rs r;stack_close_v2rs    0h..0$ #\\ 11 ndF,g n(88vD2!QX *  & &- ##At,,  # #    * t''  # #   rzraw_ops.StackCloseV2ctj|tj}|g}d}t j dd||||}d}|S)Ns StackCloseV2rrrrrr,rIrrs r;rrsM  ! !&'*:*: ;&, &   _a #)s ?' ' .r=) TV_StackPop_elem_typerrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rctjxstj}|j}|jr t dt j |d}tjd|||\}}}}|dd}t jr7d|jdf} |j} t jd| | ||\}|S)zDeprecated, use StackPopV2. Args: handle: A `Tensor` of type mutable `string`. elem_type: A `tf.DType`. name: A name for the operation (optional). Returns: A `Tensor` of type `elem_type`. Estack_pop op does not support eager execution. Arg 'handle' is a ref.rStackPopr'rr.Nre) r'rr.r6r7r8r9r:rNrOrPs r; stack_poprs    0h..0$ #\\ ^ __  K8)'886YTC!QX QK' ""$3--k: ;F::L L&'3 (' .r=zraw_ops.StackPopctd)Nrr?)r'rr.r@s r;stack_pop_eager_fallbackrs\]]r=) TV_StackPopV2_elem_typerrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rctjxstj}|j}|jr t j |d||d|}|Stj|d}tj d|||\}}}} | dd}tj"r7d|j%df} |j&} tj(d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||S#t j$rYwxYw)aPop the element at the top of the stack. Args: handle: A `Tensor` of type `resource`. The handle to a stack. elem_type: A `tf.DType`. The type of the elem that is popped. name: A name for the operation (optional). Returns: A `Tensor` of type `elem_type`. StackPopV2rN)rr.r@r)r0rr1r2rrrrrrrstack_pop_v2_eager_fallbackrrIrfr4r5rJrgrKrL) r'rr.r6r7rNrr8r9r:rOrPs r; stack_pop_v2rsS    0h..0$ #\\ 11 lD&+yBg n  K8)'88VytE!QX QK' ""$3--k: ;F::L lFG5 (' .)  & &- ##At,,  # #    ( IDd<<  # #   rKzraw_ops.StackPopV2ctj|d}tj|tj }|g}d|f}tj dd||||}tjrtjd||||\}|S)Nrs StackPopV2rrr) rIrfrrrr,rrJrL)r'rr.r@rPrOrNs r;rrs  K8)  ! !&'*:*: ;&,  #&   ]Al#)s ?' ""$ lFG5 (' .r=) TV_StackPush_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&relem swap_memoryctjxstj}|j}|jr t d|d}t j |d}tjd||||\}}}}|dd} t jrHd|jdd|jdf} |j} t jd| | | | \} | S)aDeprecated, use StackPushV2. Args: handle: A `Tensor` of type mutable `string`. elem: A `Tensor`. swap_memory: An optional `bool`. Defaults to `False`. name: A name for the operation (optional). Returns: A `Tensor`. Has the same type as `elem`. Fstack_push op does not support eager execution. Arg 'handle' is a ref.NFr StackPushr'rrr.r)r0rr1r2r3rIrr4r5rJrgrrKrL) r'rrr.r6r7r8r9r:rNrOrPs r; stack_pushrs    0h..0$ #\\ _ ``K""; >+'88F; !QX QK' ""$3%%c*M  /1F::L \674 (' .r=zraw_ops.StackPushctd)Nrr?)r'rrr.r@s r;stack_push_eager_fallbackr7rr=) TV_StackPushV2_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr t j |d|||d|}|S|d}tj|d}tj d||||\}}} } | dd}tj"rHd| j%dd| j'df} | j(} tj*d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||||S#t j$rYwxYw)ayPush an element onto the stack. Args: handle: A `Tensor` of type `resource`. The handle to a stack. elem: A `Tensor`. The tensor to be pushed onto the stack. swap_memory: An optional `bool`. Defaults to `False`. Swap `elem` to CPU. Default to false. name: A name for the operation (optional). Returns: A `Tensor`. Has the same type as `elem`. StackPushV2rN)rr.r@Frr)r0rr1r2rrrrrrrstack_push_v2_eager_fallbackrrIrr4r5rJrgrrKrL) r'rrr.r6r7rNrr8r9r:rOrPs r; stack_push_v2r<sz    0h..0$ #\\ 11 mT64 Mg nK""; >+'88f4[ "!QX QK' ""$3%%c*M  /1F::L |VW6 (' .1  & &- ##At,,  # #    ) $KdFF  # #   s0C==ED++EEEE/.E/zraw_ops.StackPushV2cf|d}tj|d}tj|g|g\}\}tj|t j }||g}d|d|f}tjdd||||}tjrtjd||||\}|S)NFrrs StackPushV2rrr) rIrrrrrr,rrJrL) r'rrr.r@rrPrOrNs r;rrmsK""; >+44dVS"E'7D  ! !&'*:*: ;&$, - 5&   ^Q|#)s ?' ""$ |VW6 (' .r=) TV_StackV2_elem_typerrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rmax_sizec tjxstj}|j}|jr t j |d||d|d|}|Stj|d}|d}tj|d}t!j"d||||\}}} } | dd}tj$rHd| j'dd| j)df} | j*} tj,d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||||S#t j$rY'wxYw)a?A stack that produces elements in first-in last-out order. Args: max_size: A `Tensor` of type `int32`. The maximum size of the stack if non-negative. If negative, the stack size is unlimited. elem_type: A `tf.DType`. The type of the elements on the stack. stack_name: An optional `string`. Defaults to `""`. Overrides the name used for the temporary stack resource. Default value is the name of the 'Stack' op (which is guaranteed unique). name: A name for the operation (optional). Returns: A `Tensor` of type `resource`. StackV2rrNrrn)rrrr.)r0rr1r2rrrrrrrstack_v2_eager_fallbackrrIrfrr4r5rJrgrrKrL) rrrr.r6r7rNrr8r9r:rOrPs r;stack_v2r s    0h..0$ #\\ 11 ixign  K8)J  \:*'88H (t5!QX QK' ""$3--k:Lll<(*F::L <2 (' .5  & &- ##At,,  # #    $ iJT  # #   s0DE'EEEE//FFzraw_ops.StackV2cVtj|d}|d}tj|d}tj|t j }|g}d|d|f}tjdd||||}tjrtjd||||\}|S)NrrnrsStackV2rrr ) rIrfrrrrrrrJrL)rrrr.r@rPrOrNs r;r r s  K8)J  \:*  # #Hgmm <(, L* =&   Z= 0`. Defaults to `0`. Maximum number of elements in the Staging Area. If > 0, inserts on the container will block when the capacity is reached. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. The maximum number of bytes allowed for Tensors in the Staging Area. If > 0, inserts will block until sufficient space is available. container: An optional `string`. Defaults to `""`. If non-empty, this queue is placed in the given container. Otherwise, a default container is used. shared_name: An optional `string`. Defaults to `""`. It is necessary to match this name to the matching Unstage Op. name: A name for the operation (optional). Returns: The created Operation. StagerorMrprqN)rorMrprqr.r@rrn)rrorMrprqr.)r0rr1r2rrrrrrrstage_eager_fallbackrrIr~rr4r5) rrorMrprqr.r6r7rNrr8r9r:s r;stagers2    0h..0$ #\\ 11 gtVZ>k9m[JgnH   x 4(L""<@,I ;7)K!!+}=+'88 $+DJ!QX *5  & &- ##At,,  # #    ! 8,;TtMM  # #   s0!C--D4DD43D48E E E z raw_ops.Stagec p|d}tj|d}|d}tj|d}|d}tj|d}|d}tj|d}tj||\}}t |}d|d|d|d|d|f } tj dd|| || } d} | S) NrrorMrnrprqrsStager)rIr~rrrzr) rrorMrprqr.r@ _attr_dtypesrPrOrNs r;rrs H   x 4(L""<@,I ;7)K!!+}=+!@@M,f, ., Y { D&   XqV!$4 1' ' .r=cbtjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYEwxYwcc} w) rO StageClearrorMrrprqNrQ@Expected list for 'dtypes' argument to 'stage_clear' Op, not %r.rrnrS)r0rr1r2rrrrrrrstage_clear_eager_fallbackrryrzr{r|rIrfr~rr4r5rUs r; stage_clearrs    0h..0$ #\\  11 lD*hh Y gn FT5M *  $&, - ..8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88Vh#/9"-D:!QX *A  & &- ##At,,  # #    ',v;TtMM  # #    ?rWzraw_ops.StageClearc t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd d|| || } d} | Scc}w) NrrrrorMrnrprqs StageClearrrYrZs r;rrYs FT5M *  $&, - ..8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   ]Al#)s ?' ' .' ?r[indexctjxstj}|j}|jr% t j |d||d|d|d|d|d|} | St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d||||||| \} } } }|dd} | s| St#j.r{d| j1dd| j1dd| j3dd| j3dd| j3df }| j4}t#j6d||| | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t||||||||S#t j$rYwxYwcc} w) aOp peeks at the values at the specified index. If the underlying container does not contain sufficient elements this op will block until it does. This Op is optimized for performance. Args: index: A `Tensor` of type `int32`. dtypes: A list of `tf.DTypes` that has length `>= 1`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `dtypes`. StagePeekrorMrrprqNrQ?Expected list for 'dtypes' argument to 'stage_peek' Op, not %r.rrn)rrrorMrprqr.)r0rr1r2rrrrrrrstage_peek_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrL)rrrorMrprqr.r6r7rNrrr8r9r:rOrPs r; stage_peekrtsm&    0h..0$ #\\  11 k4 Hnh Y gn FT5M *  #%+ , --8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'885(".)!,49!QX QK'  J ""$#++J7/ll8$k3<< 3LS\\-8:F::L \674 .W  & &- ##At,,  # #    & (f;TtMM  # #    ?s6#G IH"G==HHH--IIzraw_ops.StagePeekc tt|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}tj|tj}|g} d|d|d|d|d|f } t jd t|| | || } t jrt jd | | | | Scc}w) NrrrrorMrnrprqs StagePeekrr)ryrzr{r|rIrfr~rrrrrrrrJrL) rrrorMrprqr.r@rrPrOrNs r;rrsH FT5M *  #%+ , --8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+   6%, .,+y- >&   \3v;|#)s ?' ""$ \674 .- ?sD5ctjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | dd}t#j.r{d| j1dd| j1dd| j3dd| j3dd| j3df }| j4}t#j6d||||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} w) rv StageSizerorMrrprqNrQ?Expected list for 'dtypes' argument to 'stage_size' Op, not %r.rrnrS)r0rr1r2rrrrrrrstage_size_eager_fallbackrryrzr{r|rIrfr~rr4r5rJrrrKrLras r; stage_sizer$sf    0h..0$ #\\  11 k4X~h Y gn FT5M *  #%+ , --8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88FX".)!,49!QX QK' ""$#++J7/ll8$k3<< 3LS\\-8:F::L \674 (' .U  & &- ##At,,  # #    &,v;TtMM  # #    ?rczraw_ops.StageSizec  t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd d || || } t jrt jd || | | \} | Scc}w) Nr"rrrorMrnrprqs StageSizerrr!rerZs r;r#r#$s3 FT5M *  #%+ , --8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   \1\#)s ?' ""$ \674 (' .- ?rf) TV_TensorArray_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rsize dynamic_sizeclear_after_readtensor_array_namec tjxstj}|j}|jr t dt j |d}|d}t j|d}|d}t j|d}|d}t j|d }|d}t j|d }tjd ||||||| \} } } } | dd} t jr{d| jdd| jdd| jdd | jd d | jd f } | j }t j"d || | | \} | S) aTODO: add doc. Args: size: A `Tensor` of type `int32`. dtype: A `tf.DType`. dynamic_size: An optional `bool`. Defaults to `False`. clear_after_read: An optional `bool`. Defaults to `True`. tensor_array_name: An optional `string`. Defaults to `""`. element_shape: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Htensor_array op does not support eager execution. Arg 'handle' is a ref.r`NFr(Tr)rnr* element_shape TensorArray)r'r`r(r)r*r-r.)r0rr1r2r3rIrfrrr}r4r5rJrgrrrKrL)r'r`r(r)r*r-r.r6r7r8r9r:rNrOrPs r; tensor_arrayr/Ds    0h..0$ #\\ a bb   UG ,%L##L.A,''(8:LM''(9;NOM%%m_E-'88DL(8):%2 ?!QX QK' ""$s))'2N  02D  !346Ill./ll?+ -F ::L |VW6 (' .r=zraw_ops.TensorArrayctd)Nr,r?)r'r`r(r)r*r-r.r@s r;tensor_array_eager_fallbackr1zs_``r=ctjxstj}|j}|jr t dt j d||\}}}}|S)zTODO: add doc. Args: handle: A `Tensor` of type mutable `string`. name: A name for the operation (optional). Returns: The created Operation. Ntensor_array_close op does not support eager execution. Arg 'handle' is a ref.TensorArrayCloserFr/rs r;tensor_array_closer5}s`    0h..0$ #\\ g hh'8866!QX *r=zraw_ops.TensorArrayClosectd)Nr3r?rSs r;!tensor_array_close_eager_fallbackr7rr=ctjxstj}|j}|jr t j |d||}|Stjd||\}}}}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||S#t j$rYwxYw)zDeprecated. Use TensorArrayCloseV3 Args: handle: A `Tensor` of type `string`. name: A name for the operation (optional). Returns: The created Operation. TensorArrayCloseV2NrrF)r0rr1r2rrrrrrr$tensor_array_close_v2_eager_fallbackrr4r5rs r;tensor_array_close_v2r;s    0h..0$ #\\ 11 "D&2g n(88V$8!QX *  & &- ##At,,  # #    1 t''  # #   rzraw_ops.TensorArrayCloseV2ctj|tj}|g}d}t j dd||||}d}|S)NsTensorArrayCloseV2rrrrs r;r:r:sL  ! !&'.. 9&, &   2Al#)s ?' ' .r=ctjxstj}|j}|jr t j |d||}|Stjd||\}}}}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||S#t j$rYwxYw)ahDelete the TensorArray from its resource container. This enables the user to close and release the resource in the middle of a step/run. Args: handle: A `Tensor` of type `resource`. The handle to a TensorArray (output of TensorArray or TensorArrayGrad). name: A name for the operation (optional). Returns: The created Operation. TensorArrayCloseV3NrrF)r0rr1r2rrrrrrr$tensor_array_close_v3_eager_fallbackrr4r5rs r;tensor_array_close_v3r@s    0h..0$ #\\ 11 "D&2g n(88V$8!QX *  & &- ##At,,  # #    1 t''  # #   rzraw_ops.TensorArrayCloseV3ctj|tj}|g}d}t j dd||||}d}|S)NsTensorArrayCloseV3rrrrs r;r?r?sN  ! !&'*:*: ;&, &   2Al#)s ?' ' .r=TensorArrayConcatlengths) TV_TensorArrayConcat_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rflow_inc$tjxstj}|j}|jr t dt j |d}|d}t j|d}tjd|||||\}}}} | dd} t jrHd|jdd|jdf} |j} t jd| | | tj!| } | S)aTODO: add doc. Args: handle: A `Tensor` of type mutable `string`. flow_in: A `Tensor` of type `float32`. dtype: A `tf.DType`. element_shape_except0: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (value, lengths). value: A `Tensor` of type `dtype`. lengths: A `Tensor` of type `int64`. Otensor_array_concat op does not support eager execution. Arg 'handle' is a ref.r`Nelement_shape_except0rBr'rEr`rHr.)r0rr1r2r3rIrfr}r4r5rJrgrrKrL_TensorArrayConcatOutputr) r'rEr`rHr.r6r7r8r9r:rNrOrPs r;tensor_array_concatrKs     0h..0$ #\\ h ii   UG ,%" "--.CE\]'88FG53H"&(!QX QK' ""$s))'24Kll235F::L \67< $ * *7 3' .r=zraw_ops.TensorArrayConcatctd)NrGr?)r'rEr`rHr.r@s r;"tensor_array_concat_eager_fallbackrMrr=TensorArrayConcatV2) TV_TensorArrayConcatV2_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc dtjxstj}|j}|jr5 t j |d|||d|d| }t j|}|Stj |d}|d}tj"|d}t%j&d|||||\} } } } | dd}tj(rHd| j+dd| j-df} | j.} tj0d| | |t j|}|S#tj$r }tj||Yd}~nd}~wtj$rYnwxYw t||||||S#tj$rY:wxYw)aDeprecated. Use TensorArrayConcatV3 Args: handle: A `Tensor` of type `string`. flow_in: A `Tensor` of type `float32`. dtype: A `tf.DType`. element_shape_except0: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (value, lengths). value: A `Tensor` of type `dtype`. lengths: A `Tensor` of type `int64`. rNr`rHNr`rHr.r@rI)r0rr1r2rr_TensorArrayConcatV2Outputrrrrrr%tensor_array_concat_v2_eager_fallbackrrIrfr}r4r5rJrgrrKrLr'rEr`rHr.r6r7rNrr8r9r:rOrPs r;tensor_array_concat_v2rU(s    0h..0$ #\\  11 #T67GU!68g+009g n   UG ,%" "--.CE\]'88fgU5J$(*!QX QK' ""$s))'24Kll235F::L |VW> & , ,W 5' .7  & &- ##At,,  # #    2 ' 5DdLL  # #   03D<<FE**FFFF/.F/zraw_ops.TensorArrayConcatV2ctj|d}|d}tj|d}tj|t j }tj|t j}||g}d|d|f}tjdd||||}tjrtjd|||tj|}|S)Nr`rHsTensorArrayConcatV2rrrN) rIrfr}rrrrrrrJrLrRr r'rEr`rHr.r@rPrOrNs r;rSrSas   UG ,%" "--.CE\]  ! !&'.. 9&  " "7GOO <''", U35J K&   3Q|#)s ?' ""$ |VW> & , ,W 5' .r=TensorArrayConcatV3) TV_TensorArrayConcatV3_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc dtjxstj}|j}|jr5 t j |d|||d|d| }t j|}|Stj |d}|d}tj"|d}t%j&d|||||\} } } } | dd}tj(rHd| j+dd| j-df} | j.} tj0d| | |t j|}|S#tj$r }tj||Yd}~nd}~wtj$rYnwxYw t||||||S#tj$rY:wxYw)aConcat the elements from the TensorArray into value `value`. Takes `T` elements of shapes ``` (n0 x d0 x d1 x ...), (n1 x d0 x d1 x ...), ..., (n(T-1) x d0 x d1 x ...) ``` and concatenates them into a Tensor of shape: ``` (n0 + n1 + ... + n(T-1) x d0 x d1 x ...) ``` All elements must have the same shape (excepting the first dimension). Args: handle: A `Tensor` of type `resource`. The handle to a TensorArray. flow_in: A `Tensor` of type `float32`. A float scalar that enforces proper chaining of operations. dtype: A `tf.DType`. The type of the elem that is returned. element_shape_except0: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. The expected shape of an element, if known, excluding the first dimension. Used to validate the shapes of TensorArray elements. If this shape is not fully specified, concatenating zero-size TensorArrays is an error. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (value, lengths). value: A `Tensor` of type `dtype`. lengths: A `Tensor` of type `int64`. rYr`rHNrQrI)r0rr1r2rr_TensorArrayConcatV3Outputrrrrrr%tensor_array_concat_v3_eager_fallbackrrIrfr}r4r5rJrgrrKrLrTs r;tensor_array_concat_v3r^ysF    0h..0$ #\\  11 #T67GU!68g+009g n   UG ,%" "--.CE\]'88fgU5J$(*!QX QK' ""$s))'24Kll235F::L |VW> & , ,W 5' .7  & &- ##At,,  # #    2 ' 5DdLL  # #   rVzraw_ops.TensorArrayConcatV3ctj|d}|d}tj|d}tj|t j }tj|t j}||g}d|d|f}tjdd||||}tjrtjd|||tj|}|S)Nr`rHsTensorArrayConcatV3rrrY) rIrfr}rrrr,rrrJrLr\rrXs r;r]r]s   UG ,%" "--.CE\]  ! !&'*:*: ;&  " "7GOO <''", U35J K&   3Q|#)s ?' ""$ |VW> & , ,W 5' .r=) TV_TensorArrayGather_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr t dt j |d}|d}t j|d}tjd||||||\}}} } | dd} t jrHd| jdd| jdf} | j} t jd| | | | \} | S)ajTODO: add doc. Args: handle: A `Tensor` of type mutable `string`. indices: A `Tensor` of type `int32`. flow_in: A `Tensor` of type `float32`. dtype: A `tf.DType`. element_shape: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. name: A name for the operation (optional). Returns: A `Tensor` of type `dtype`. Otensor_array_gather op does not support eager execution. Arg 'handle' is a ref.r`Nr-TensorArrayGatherr'rrEr`r-r.r0rr1r2r3rIrfr}r4r5rJrgrrKrL)r'rrEr`r-r.r6r7r8r9r:rNrOrPs r;tensor_array_gatherrfs    0h..0$ #\\ h ii   UG ,%M%%m_E-'88FGW#( "&(!QX QK' ""$s))'2Oll?+-F::L \67< (' .r=zraw_ops.TensorArrayGatherctd)Nrbr?)r'rrEr`r-r.r@s r;"tensor_array_gather_eager_fallbackrhrr=) TV_TensorArrayGatherV2_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr! t j |d||||d|d| }|Stj|d}|d}tj|d}t!j"d||||||\} } } } | dd}tj$rHd| j'dd| j)df} | j*}tj,d|| ||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rY+wxYw)awDeprecated. Use TensorArrayGatherV3 Args: handle: A `Tensor` of type `string`. indices: A `Tensor` of type `int32`. flow_in: A `Tensor` of type `float32`. dtype: A `tf.DType`. element_shape: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. name: A name for the operation (optional). Returns: A `Tensor` of type `dtype`. TensorArrayGatherV2r`r-Nr`r-r.r@rd)r0rr1r2rrrrrrr%tensor_array_gather_v2_eager_fallbackrrIrfr}r4r5rJrgrrKrLr'rrEr`r-r.r6r7rNrr8r9r:rOrPs r;tensor_array_gather_v2ros    0h..0$ #\\ 11 #T67GW  /gn   UG ,%M%%m_E-'88fg'.e-:G!QX QK' ""$s))'2Oll?+-F::L |VW> (' .7  & &- ##At,,  # #    2 '7%}  # #   0DE+EEE#E55F  F zraw_ops.TensorArrayGatherV2ctj|d}|d}tj|d}tj|t j }tj|t j}tj|t j}|||g}d|d|f}tjdd||||} tjrtjd||| | \} | S)Nr`r-sTensorArrayGatherV2rrrk) rIrfr}rrrrrrrrJrL r'rrEr`r-r.r@rPrOrNs r;rmrm<s   UG ,%M%%m_E-  ! !&'.. 9&  " "7GMM :'  " "7GOO <''7+, UO] ;&   3Q|#)s ?' ""$ |VW> (' .r=) TV_TensorArrayGatherV3_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr! t j |d||||d|d| }|Stj|d}|d}tj|d}t!j"d||||||\} } } } | dd}tj$rHd| j'dd| j)df} | j*}tj,d|| ||\}|S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rY+wxYw)apGather specific elements from the TensorArray into output `value`. All elements selected by `indices` must have the same shape. Args: handle: A `Tensor` of type `resource`. The handle to a TensorArray. indices: A `Tensor` of type `int32`. The locations in the TensorArray from which to read tensor elements. flow_in: A `Tensor` of type `float32`. A float scalar that enforces proper chaining of operations. dtype: A `tf.DType`. The type of the elem that is returned. element_shape: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. The expected shape of an element, if known. Used to validate the shapes of TensorArray elements. If this shape is not fully specified, gathering zero-size TensorArrays is an error. name: A name for the operation (optional). Returns: A `Tensor` of type `dtype`. TensorArrayGatherV3r`r-Nrlrd)r0rr1r2rrrrrrr%tensor_array_gather_v3_eager_fallbackrrIrfr}r4r5rJrgrrKrLrns r;tensor_array_gather_v3rwQs*    0h..0$ #\\ 11 #T67GW  /gn   UG ,%M%%m_E-'88fg'.e-:G!QX QK' ""$s))'2Oll?+-F::L |VW> (' .7  & &- ##At,,  # #    2 '7%}  # #   rpzraw_ops.TensorArrayGatherV3ctj|d}|d}tj|d}tj|t j }tj|t j}tj|t j}|||g}d|d|f}tjdd||||} tjrtjd||| | \} | S)Nr`r-sTensorArrayGatherV3rrru) rIrfr}rrrr,rrrrJrLrrs r;rvrvs   UG ,%M%%m_E-  ! !&'*:*: ;&  " "7GMM :'  " "7GOO <''7+, UO] ;&   3Q|#)s ?' ""$ |VW> (' .r=sourcectjxstj}|j}|jr t dt j |d}tjd||||\}}}}|dd} t jr7d|jdf} |j} t jd| | | | \} | S)zTODO: add doc. Args: handle: A `Tensor` of type `string`. flow_in: A `Tensor` of type `float32`. source: A `string`. name: A name for the operation (optional). Returns: A `Tensor` of type mutable `string`. Rtensor_array_grad op does not support eager execution. Arg 'grad_handle' is a ref.ryTensorArrayGradr'rEryr.N) r0rr1r2r3rIrr4r5rJrrKrL) r'rEryr.r6r7r8r9r:rNrOrPs r;tensor_array_gradr~s    0h..0$ #\\ k ll   VX .&'88&'& $&!QX QK' ""$ X. /F::L <: (' .r=zraw_ops.TensorArrayGradctd)Nr{r?)r'rEryr.r@s r; tensor_array_grad_eager_fallbackrsijjr=c tjxstj}|j}|jr t j |d|||d|}|Stj|d}tj d||||\}}} } | dd}tj"r7d| j%df} | j&} tj(d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t|||||S#t j$rYwxYw)zDeprecated. Use TensorArrayGradV3 Args: handle: A `Tensor` of type `string`. flow_in: A `Tensor` of type `float32`. source: A `string`. name: A name for the operation (optional). Returns: A `Tensor` of type `string`. TensorArrayGradV2ryNryr.r@r})r0rr1r2rrrrrrr#tensor_array_grad_v2_eager_fallbackrrIrr4r5rJrrKrL r'rEryr.r6r7rNrr8r9r:rOrPs r;tensor_array_grad_v2rsZ    0h..0$ #\\ 11 !4(FLg n   VX .&'88FGF"&(!QX QK' ""$ X. /F::L \67< (' .+  & &- ##At,,  # #    0 '&t??  # #   rzraw_ops.TensorArrayGradV2chtj|d}tj|tj }tj|tj }||g}d|f}tjdd||||}tjrtjd||||\}|S)NrysTensorArrayGradV2rrr) rIrrrrrrrrJrLr'rEryr.r@rPrOrNs r;rrs   VX .&  ! !&'.. 9&  " "7GOO <''", f &   11\#)s ?' ""$ \67< (' .r=TensorArrayGradV3 grad_handleflow_outc tjxstj}|j}|jr3 t j |d|||d|}t j|}|Stj |d}t#j$d||||\}}} } | dd}tj&r7d| j)df} | j*} tj,d| | |t j|}|S#tj$r }tj||Yd}~nd}~wtj$rYnwxYw t|||||S#tj$rY wxYw)a- Creates a TensorArray for storing the gradients of values in the given handle. If the given TensorArray gradient already exists, returns a reference to it. Locks the size of the original TensorArray by disabling its dynamic size flag. **A note about the input flow_in:** The handle flow_in forces the execution of the gradient lookup to occur only after certain other operations have occurred. For example, when the forward TensorArray is dynamically sized, writes to this TensorArray may resize the object. The gradient TensorArray is statically sized based on the size of the forward TensorArray when this operation executes. Furthermore, the size of the forward TensorArray is frozen by this call. As a result, the flow is used to ensure that the call to generate the gradient TensorArray only happens after all writes are executed. In the case of dynamically sized TensorArrays, gradient computation should only be performed on read operations that have themselves been chained via flow to occur only after all writes have executed. That way the final size of the forward TensorArray is known when this operation is called. **A note about the source attribute:** TensorArray gradient calls use an accumulator TensorArray object. If multiple gradients are calculated and run in the same session, the multiple gradient nodes may accidentally flow through the same accumulator TensorArray. This double counts and generally breaks the TensorArray gradient flow. The solution is to identify which gradient call this particular TensorArray gradient is being called in. This is performed by identifying a unique string (e.g. "gradients", "gradients_1", ...) from the input gradient Tensor's name. This string is used as a suffix when creating the TensorArray gradient object here (the attribute `source`). The attribute `source` is added as a suffix to the forward TensorArray's name when performing the creation / lookup, so that each separate gradient calculation gets its own TensorArray accumulator. Args: handle: A `Tensor` of type `resource`. The handle to the forward TensorArray. flow_in: A `Tensor` of type `float32`. A float scalar that enforces proper chaining of operations. source: A `string`. The gradient source string, used to decide which gradient TensorArray to return. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (grad_handle, flow_out). grad_handle: A `Tensor` of type `resource`. flow_out: A `Tensor` of type `float32`. rryNrr})r0rr1r2rr_TensorArrayGradV3Outputrrrrrr#tensor_array_grad_v3_eager_fallbackrrIrr4r5rJrrKrLrs r;tensor_array_grad_v3rstp    0h..0$ #\\ 11 !4(FLg(..w7g n   VX .&'88FGF"&(!QX QK' ""$ X. /F::L \67< $ * *7 3' .+  & &- ##At,,  # #    0 '&t??  # #   s01DE!D<<EEE))F?Fzraw_ops.TensorArrayGradV3ctj|d}tj|tj }tj|tj }||g}d|f}tjdd||||}tjrtjd|||tj|}|S)NrysTensorArrayGradV3rrr) rIrrrrr,rrrJrLrrrs r;rr_s   VX .&  ! !&'*:*: ;&  " "7GOO <''", f &   11\#)s ?' ""$ \67< $ * *7 3' .r=TensorArrayGradWithShapeshape_to_prependc  tjxstj}|j}|jr4 t j |d||||d|}t j|}|Stj |d}t#j$d|||||\} } } } | dd}tj&r7d| j)df} | j*} tj,d| | |t j|}|S#tj$r }tj||Yd}~nd}~wtj$rYnwxYw t||||||S#tj$rYwxYw)aCreates a TensorArray for storing multiple gradients of values in the given handle. Similar to TensorArrayGradV3. However it creates an accumulator with an expanded shape compared to the input TensorArray whose gradient is being computed. This enables multiple gradients for the same TensorArray to be calculated using the same accumulator. Args: handle: A `Tensor` of type `resource`. The handle to the forward TensorArray. flow_in: A `Tensor` of type `float32`. A float scalar that enforces proper chaining of operations. shape_to_prepend: A `Tensor` of type `int32`. An int32 vector representing a shape. Elements in the gradient accumulator will have shape which is this shape_to_prepend value concatenated with shape of the elements in the TensorArray corresponding to the input handle. source: A `string`. The gradient source string, used to decide which gradient TensorArray to return. name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (grad_handle, flow_out). grad_handle: A `Tensor` of type `resource`. flow_out: A `Tensor` of type `float32`. rryNr)r'rErryr.)r0rr1r2rr_TensorArrayGradWithShapeOutputrrrrrr+tensor_array_grad_with_shape_eager_fallbackrrIrr4r5rJrrKrL)r'rErryr.r6r7rNrr8r9r:rOrPs r;tensor_array_grad_with_shaperrs8    0h..0$ #\\  11 ($(F,g055g>g n   VX .&'88"675E+1>!QX QK' ""$ X. /F::L "L&'C + 1 1' :' ./  & &- ##At,,  # #    8 '+F  # #   s02DE#D>>EEE,,FFz raw_ops.TensorArrayGradWithShapectj|d}tj|tj }tj|tj }tj|tj}|||g}d|f}tjdd||||}tjrtjd|||tj|}|S)NrysTensorArrayGradWithShaperrr) rIrrrrr,rrrrJrLrr) r'rErryr.r@rPrOrNs r;rrs   VX .&  ! !&'*:*: ;&  " "7GOO <'++,&   4a #)s ?' ""$  fg? (' .r=) TV_TensorArrayScatterV3_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr t j |d|||||}|Stjd|||||\} } } } | dd}tj r7d| j#df} | j$} tj&d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||||S#t j$rYwxYw)afScatter the data from the input value into specific TensorArray elements. `indices` must be a vector, its length must match the first dim of `value`. Args: handle: A `Tensor` of type `resource`. The handle to a TensorArray. indices: A `Tensor` of type `int32`. The locations at which to write the tensor elements. value: A `Tensor`. The concatenated tensor to write to the TensorArray. flow_in: A `Tensor` of type `float32`. A float scalar that enforces proper chaining of operations. name: A name for the operation (optional). Returns: A `Tensor` of type `float32`. TensorArrayScatterV3Nrrr)r0rr1r2rrrrrrr&tensor_array_scatter_v3_eager_fallbackrr4r5rIrJrgrKrLrs r;tensor_array_scatter_v3rsP"    0h..0$ #\\ 11 $dFGUGMg n(88vwe(/d<!QX QK' ""$3%%c* +F::L  fg? (' .)  & &- ##At,,  # #    3 '5'$@@  # #   rzraw_ops.TensorArrayScatterV3ctj|g|g\}\}tj|tj }tj|tj }tj|tj}||||g}d|f}tjdd||||} tjrtjd||| | \} | S)NrsTensorArrayScatterV3rrr rIrrrrr,rrrrJrLrs r;rr1s55ugsBG'8E  ! !&'*:*: ;&  " "7GMM :'  " "7GOO <''5'2, >&   4a #)s ?' ""$  fg? (' .r=cZtjxstj}|j}|jr t dt j d|||\}}}}|dd}tjr&d} |j} tjd| | ||\}|S)zTODO: add doc. Args: handle: A `Tensor` of type mutable `string`. flow_in: A `Tensor` of type `float32`. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. Mtensor_array_size op does not support eager execution. Arg 'handle' is a ref.TensorArraySizer'rEr.NrGrH) r'rEr.r6r7r8r9r:rNrOrPs r;tensor_array_sizerAs    0h..0$ #\\ f gg'88&'F!QX QK' ""$ F::L <: (' .r=zraw_ops.TensorArraySizectd)Nrr?)r'rEr.r@s r; tensor_array_size_eager_fallbackr_rr=cbtjxstj}|j}|jr t j |d|||}|Stjd|||\}}}} | dd}tj r&d} |j"} tj$d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||S#t j$rYwxYw)zDeprecated. Use TensorArraySizeV3 Args: handle: A `Tensor` of type `string`. flow_in: A `Tensor` of type `float32`. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. TensorArraySizeV2NrrrG)r0rr1r2rrrrrrr#tensor_array_size_v2_eager_fallbackrr4r5rIrJrKrL r'rEr.r6r7rNrr8r9r:rOrPs r;tensor_array_size_v2rbs/    0h..0$ #\\ 11 !4:g n(88FG$H!QX QK' ""$ F::L \67< (' .'  & &- ##At,,  # #    0 '$00  # #   0B>>DC,,DD DD.-D.zraw_ops.TensorArraySizeV2c8tj|tj}tj|tj}||g}d}t j dd||||}t jrt jd||||\}|S)NsTensorArraySizeV2rrr) rrrrrrIrrJrLr'rEr.r@rPrOrNs r;rrs  ! !&'.. 9&  " "7GOO <''", &   11\#)s ?' ""$ \67< (' .r=cbtjxstj}|j}|jr t j |d|||}|Stjd|||\}}}} | dd}tj r&d} |j"} tj$d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||S#t j$rYwxYw)auGet the current size of the TensorArray. Args: handle: A `Tensor` of type `resource`. The handle to a TensorArray (output of TensorArray or TensorArrayGrad). flow_in: A `Tensor` of type `float32`. A float scalar that enforces proper chaining of operations. name: A name for the operation (optional). Returns: A `Tensor` of type `int32`. TensorArraySizeV3NrrrG)r0rr1r2rrrrrrr#tensor_array_size_v3_eager_fallbackrr4r5rIrJrKrLrs r;tensor_array_size_v3rs/    0h..0$ #\\ 11 !4:g n(88FG$H!QX QK' ""$ F::L \67< (' .'  & &- ##At,,  # #    0 '$00  # #   rzraw_ops.TensorArraySizeV3c8tj|tj}tj|tj}||g}d}t j dd||||}t jrt jd||||\}|S)NsTensorArraySizeV3rrr) rrrr,rrIrrJrLrs r;rrs  ! !&'*:*: ;&  " "7GOO <''", &   11\#)s ?' ""$ \67< (' .r=) TV_TensorArraySplit_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rctjxstj}|j}|jr t dt j d|||||\}}}} | dd} tjr7d|jdf} |j} tjd| | | | \} | S)aTODO: add doc. Args: handle: A `Tensor` of type mutable `string`. value: A `Tensor`. lengths: A `Tensor` of type `int64`. flow_in: A `Tensor` of type `float32`. name: A name for the operation (optional). Returns: A `Tensor` of type `float32`. Ntensor_array_split op does not support eager execution. Arg 'handle' is a ref.TensorArraySplitr'r4rCrEr.Nrr) r'r4rCrEr.r6r7r8r9r:rNrOrPs r;tensor_array_splitrs    0h..0$ #\\ g hh'886$+$8!QX QK' ""$3%%c* +F::L L&'; (' .r=zraw_ops.TensorArraySplitctd)Nrr?)r'r4rCrEr.r@s r;!tensor_array_split_eager_fallbackrrr=) TV_TensorArraySplitV2_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr t j |d|||||}|Stjd|||||\} } } } | dd}tj r7d| j#df} | j$} tj&d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||||S#t j$rYwxYw)aDeprecated. Use TensorArraySplitV3 Args: handle: A `Tensor` of type `string`. value: A `Tensor`. lengths: A `Tensor` of type `int64`. flow_in: A `Tensor` of type `float32`. name: A name for the operation (optional). Returns: A `Tensor` of type `float32`. TensorArraySplitV2Nrrr)r0rr1r2rrrrrrr$tensor_array_split_v2_eager_fallbackrr4r5rIrJrgrKrLr'r4rCrEr.r6r7rNrr8r9r:rOrPs r;tensor_array_split_v2rsP    0h..0$ #\\ 11 "D&%'Kg n(88V5'&-D:!QX QK' ""$3%%c* +F::L lFG= (' .)  & &- ##At,,  # #    1 %'$@@  # #   rzraw_ops.TensorArraySplitV2ctj|g|g\}\}tj|tj }tj|tj }tj|tj}||||g}d|f}tjdd||||} tjrtjd||| | \} | S)NrsTensorArraySplitV2rrr) rIrrrrrrrrrrJrL r'r4rCrEr.r@rrPrOrNs r;rr)s55ugsBG'8E  ! !&'.. 9&  " "7GMM :'  " "7GOO <'%'2, >&   2Al#)s ?' ""$ lFG= (' .r=) TV_TensorArraySplitV3_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr t j |d|||||}|Stjd|||||\} } } } | dd}tj r7d| j#df} | j$} tj&d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||||S#t j$rYwxYw)aSplit the data from the input value into TensorArray elements. Assuming that `lengths` takes on values ``` (n0, n1, ..., n(T-1)) ``` and that `value` has shape ``` (n0 + n1 + ... + n(T-1) x d0 x d1 x ...), ``` this splits values into a TensorArray with T tensors. TensorArray index t will be the subtensor of values with starting position ``` (n0 + n1 + ... + n(t-1), 0, 0, ...) ``` and having size ``` nt x d0 x d1 x ... ``` Args: handle: A `Tensor` of type `resource`. The handle to a TensorArray. value: A `Tensor`. The concatenated tensor to write to the TensorArray. lengths: A `Tensor` of type `int64`. The vector of lengths, how to split the rows of value into the TensorArray. flow_in: A `Tensor` of type `float32`. A float scalar that enforces proper chaining of operations. name: A name for the operation (optional). Returns: A `Tensor` of type `float32`. TensorArraySplitV3Nrrr)r0rr1r2rrrrrrr$tensor_array_split_v3_eager_fallbackrr4r5rIrJrgrKrLrs r;tensor_array_split_v3r;sQT    0h..0$ #\\ 11 "D&%'Kg n(88V5'&-D:!QX QK' ""$3%%c* +F::L lFG= (' .)  & &- ##At,,  # #    1 %'$@@  # #   rzraw_ops.TensorArraySplitV3ctj|g|g\}\}tj|tj }tj|tj }tj|tj}||||g}d|f}tjdd||||} tjrtjd||| | \} | S)NrsTensorArraySplitV3rrr) rIrrrrr,rrrrrJrLrs r;rrs55ugsBG'8E  ! !&'*:*: ;&  " "7GMM :'  " "7GOO <'%'2, >&   2Al#)s ?' ""$ lFG= (' .r=) TV_TensorArrayUnpack_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc~tjxstj}|j}|jr t dt j d||||\}}}}|dd} tjr7d|jdf} |j} tjd| | | | \} | S)zTODO: add doc. Args: handle: A `Tensor` of type mutable `string`. value: A `Tensor`. flow_in: A `Tensor` of type `float32`. name: A name for the operation (optional). Returns: A `Tensor` of type `float32`. Otensor_array_unpack op does not support eager execution. Arg 'handle' is a ref.TensorArrayUnpack)r'r4rEr.Nrr) r'r4rEr.r6r7r8r9r:rNrOrPs r;tensor_array_unpackrs    0h..0$ #\\ h ii'88F%"&(!QX QK' ""$3%%c* +F::L \67< (' .r=zraw_ops.TensorArrayUnpackctd)Nrr?)r'r4rEr.r@s r;"tensor_array_unpack_eager_fallbackrrr=) TV_TensorArrayV2_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc,tjxstj}|j}|jr% t j |d||d|d|d|d|d|} | Stj|d}|d}tj|d}|d }tj |d}|d }tj |d}|d }tj"|d}t%j&d||||||| \} } } } | dd} tj(r{d| j+dd| j-dd| j/dd| j/dd| j-df }| j0}tj2d||| | \} | S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t||||||||S#t j$rYwxYw) aDeprecated. Use TensorArrayV3 Args: size: A `Tensor` of type `int32`. dtype: A `tf.DType`. element_shape: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. dynamic_size: An optional `bool`. Defaults to `False`. clear_after_read: An optional `bool`. Defaults to `True`. tensor_array_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A `Tensor` of type `string`. TensorArrayV2r`r-r(r)r*N)r`r-r(r)r*r.r@FTrn)r'r`r-r(r)r*r.)r0rr1r2rrrrrrrtensor_array_v2_eager_fallbackrrIrfr}rrr4r5rJrgrrrKrL)r'r`r-r(r)r*r.r6r7rNrr8r9r:rOrPs r;tensor_array_v2rsG    0h..0$ #\\  11 otT7E?~|5G-/@Bgn   UG ,%M%%m_E-L##L.A,''(8:LM''(9;NO'88d%}&2*:+<4 I!QX QK' ""$s))'2Oll?+^  02D  !346Ill./ 1F ::L vw8 (' .S  & &- ##At,,  # #    + e=#6F-DdDD  # #   s0#FG%1G  G%$G%)G<<HHzraw_ops.TensorArrayV2c tj|d}|d}tj|d}|d}tj|d}|d}tj|d}|d}tj|d}t j |tj}|g}d|d|d|d|d|f } tjd d || || } tjrtjd || | | \} | S) Nr`r-Fr(Tr)rnr*s TensorArrayV2rrr) rIrfr}rrrrrrrrJrL) r'r`r-r(r)r*r.r@rPrOrNs r;rrs   UG ,%M%%m_E-L##L.A,''(8:LM''(9;NO  gmm 4$, UO]N"$46I &   -q#)s ?' ""$ vw8 (' .r= TensorArrayV3flow) TV_TensorArrayV3_dtyperrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&ridentical_element_shapesctjxstj}|j} | jr< t j |d||d|d|d|d|d|d|} t j| } | Stj |d}|d}tj"|d}|d }tj$|d}|d }tj$|d}|d }tj$|d}|d }tj&|d}t)j*d|||||||| \} } } }|dd} tj,rd| j/dd| j1dd| j3dd| j3dd| j3dd| j1df }| j4}tj6d||| t j| } | S#tj$r } tj| |Yd} ~ nd} ~ wtj$rYnwxYw t||||||||| S#tj$rYwxYw)aAn array of Tensors of given size. Write data via Write and read via Read or Pack. Args: size: A `Tensor` of type `int32`. The size of the array. dtype: A `tf.DType`. The type of the elements on the tensor_array. element_shape: An optional `tf.TensorShape` or list of `ints`. Defaults to `None`. The expected shape of an element, if known. Used to validate the shapes of TensorArray elements. If this shape is not fully specified, gathering zero-size TensorArrays is an error. dynamic_size: An optional `bool`. Defaults to `False`. A boolean that determines whether writes to the TensorArray are allowed to grow the size. By default, this is not allowed. clear_after_read: An optional `bool`. Defaults to `True`. If true (default), Tensors in the TensorArray are cleared after being read. This disables multiple read semantics but allows early release of memory. identical_element_shapes: An optional `bool`. Defaults to `False`. If true (default is false), then all elements in the TensorArray will be expected to have identical shapes. This allows certain behaviors, like dynamically checking for consistent shapes on write, and being able to fill in properly shaped zero tensors on stack -- even if the element_shape attribute is not fully defined. tensor_array_name: An optional `string`. Defaults to `""`. Overrides the name used for the temporary tensor_array resource. Default value is the name of the 'TensorArray' op (which is guaranteed unique). name: A name for the operation (optional). Returns: A tuple of `Tensor` objects (handle, flow). handle: A `Tensor` of type `resource`. flow: A `Tensor` of type `float32`. rr`r-r(r)r r*N)r`r-r(r)r r*r.r@FTrn)r'r`r-r(r)r r*r.)r0rr1r2rr_TensorArrayV3Outputrrrrrrtensor_array_v3_eager_fallbackrrIrfr}rrr4r5rJrgrrrKrL)r'r`r-r(r)r r*r.r6r7rNrr8r9r:rOrPs r;tensor_array_v3r$sL    0h..0$ #\\  11 otT7E?~|5G4 "57H Jg %**73g n   UG ,%M%%m_E-L##L.A,''(8:LM%$%//0HJde''(9;NO'88d%}&2*:2J+<4 I!QX QK' ""$s))'2Oll?+^  02D  !34(  !;<!3<<0C#D FF::L vw8 & &w /' .a  & &- ##At,,  # #    + e=#6F#;-Dd DD  # #   s0:G22H9H  H98H9=II('I(zraw_ops.TensorArrayV3c Xtj|d}|d}tj|d}|d}tj|d}|d}tj|d}|d}tj|d}|d}tj|d }t j |tj}|g} d|d|d|d|d|d |f } tjd d | | || } tjrtjd | | | tj| } | S)Nr`r-Fr(Tr)r rnr*s TensorArrayV3rrr)rIrfr}rrrrrrrrJrLr r) r'r`r-r(r)r r*r.r@rPrOrNs r;r r sM   UG ,%M%%m_E-L##L.A,''(8:LM%$%//0HJde''(9;NO  gmm 4$, UO]N"$468K &   -q#)s ?' ""$ vw8 & &w /' .r=) TV_TensorArrayWrite_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rctjxstj}|j}|jr t dt j d|||||\}}}} | dd} tjr7d|jdf} |j} tjd| | | | \} | S)aTODO: add doc. Args: handle: A `Tensor` of type mutable `string`. index: A `Tensor` of type `int32`. value: A `Tensor`. flow_in: A `Tensor` of type `float32`. name: A name for the operation (optional). Returns: A `Tensor` of type `float32`. Ntensor_array_write op does not support eager execution. Arg 'handle' is a ref.TensorArrayWriter'rr4rEr.Nrr) r'rr4rEr.r6r7r8r9r:rNrOrPs r;tensor_array_writers    0h..0$ #\\ g hh'886e$+$8!QX QK' ""$3%%c* +F::L L&'; (' .r=zraw_ops.TensorArrayWritectd)Nrr?)r'rr4rEr.r@s r;!tensor_array_write_eager_fallbackrrr=) TV_TensorArrayWriteV2_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr t j |d|||||}|Stjd|||||\} } } } | dd}tj r7d| j#df} | j$} tj&d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||||S#t j$rYwxYw)aDeprecated. Use TensorArrayGradV3 Args: handle: A `Tensor` of type `string`. index: A `Tensor` of type `int32`. value: A `Tensor`. flow_in: A `Tensor` of type `float32`. name: A name for the operation (optional). Returns: A `Tensor` of type `float32`. TensorArrayWriteV2Nrrr)r0rr1r2rrrrrrr$tensor_array_write_v2_eager_fallbackrr4r5rIrJrgrKrLr'rr4rEr.r6r7rNrr8r9r:rOrPs r;tensor_array_write_v2rsN    0h..0$ #\\ 11 "D&%Ig n(88V5&-D:!QX QK' ""$3%%c* +F::L lFG= (' .)  & &- ##At,,  # #    1 %d>>  # #   rzraw_ops.TensorArrayWriteV2ctj|g|g\}\}tj|tj }tj|tj }tj|tj}||||g}d|f}tjdd||||} tjrtjd||| | \} | S)NrsTensorArrayWriteV2rrrr r'rr4rEr.r@rrPrOrNs r;rrs55ugsBG'8E  ! !&'.. 9&   6%  " "7GOO <'%0, >&   2Al#)s ?' ""$ lFG= (' .r=) TV_TensorArrayWriteV3_Trrrrrrrrrrrrrrrrrrrrr r!r"rrr#r$rr%r&rc tjxstj}|j}|jr t j |d|||||}|Stjd|||||\} } } } | dd}tj r7d| j#df} | j$} tj&d| | ||\}|S#t j$r }tj||Yd}~nd}~wt j$rYnwxYw t||||||S#t j$rYwxYw)aPush an element onto the tensor_array. Args: handle: A `Tensor` of type `resource`. The handle to a TensorArray. index: A `Tensor` of type `int32`. The position to write to inside the TensorArray. value: A `Tensor`. The tensor to write to the TensorArray. flow_in: A `Tensor` of type `float32`. A float scalar that enforces proper chaining of operations. name: A name for the operation (optional). Returns: A `Tensor` of type `float32`. TensorArrayWriteV3Nrrr)r0rr1r2rrrrrrr$tensor_array_write_v3_eager_fallbackrr4r5rIrJrgrKrLrs r;tensor_array_write_v3r$sN    0h..0$ #\\ 11 "D&%Ig n(88V5&-D:!QX QK' ""$3%%c* +F::L lFG= (' .)  & &- ##At,,  # #    1 %d>>  # #   rzraw_ops.TensorArrayWriteV3ctj|g|g\}\}tj|tj }tj|tj }tj|tj}||||g}d|f}tjdd||||} tjrtjd||| | \} | S)NrsTensorArrayWriteV3rrr"rrs r;r#r#@s55ugsBG'8E  ! !&'*:*: ;&   6%  " "7GOO <'%0, >&   2Al#)s ?' ""$ lFG= (' .r=ctjxstj}|j}|jr$ t j |d|d|d|d|d|d| }|St|ttfst!d |z|D cgc]} t#j$| d}} |d }t#j&|d}|d }t#j&|d}|d }t#j(|d}|d }t#j(|d}t+j,d|||||| \} } } } | dd}|s| St#j.r{d| j1dd| j1dd| j3dd| j3dd| j3df }| j4}t#j6d||||S#t j$r } tj| |Yd} ~ nd} ~ wt j$rYnwxYw t|||||||S#t j$rYwxYwcc} w) aSOp is similar to a lightweight Dequeue. The basic functionality is similar to dequeue with many fewer capabilities and options. This Op is optimized for performance. Args: dtypes: A list of `tf.DTypes` that has length `>= 1`. capacity: An optional `int` that is `>= 0`. Defaults to `0`. memory_limit: An optional `int` that is `>= 0`. Defaults to `0`. container: An optional `string`. Defaults to `""`. shared_name: An optional `string`. Defaults to `""`. name: A name for the operation (optional). Returns: A list of `Tensor` objects of type `dtypes`. UnstagerorMrrprqNrQ >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+'88&8 , *7!QX QK'  J ""$#++J7/ll8$k3<< 3LS\\-8:F::L <2 .W  & &- ##At,,  # #    #,v;TtMM  # #    ?rczraw_ops.Unstagec *t|ttfstd|z|Dcgc]}t j |d}}|d}t j |d}|d}t j |d}|d}t j|d}|d}t j|d}g}d|d|d|d|d|f } t jd t||| || } t jrt jd || | | Scc}w) Nr(rrrorMrnrprqsUnstagerr') ryrzr{r|rIrfr~rrrrJrLrZs r;r)r)s0 FT5M *  "( ) **8> >H  r8 , >& > H   x 4(L""<@,I ;7)K!!+}=+, .,+y- >&   ZV\#)s ?' ""$ <2 .+ ?sD)N)FN)FFrmN)rnrnrN)rrrnrnN)rmN)rrrrrrnN)rnN)FTrnNN)NN)NFTrnN)NFTFrnN(__doc__ collectionstensorflow.pythonrtensorflow.python.eagerrr0rrrrItensorflow.python.frameworkrrtensorflow.security.fuzzing.pyr_atypesr _op_def_registryr rr r4"tensorflow.python.util.deprecationr tensorflow.python.utilr r tensorflow.python.util.tf_exportrtypingrrrtyping_extensionsrrStringInt64r< to_raw_opr,rAInt32rQrErTr[rYr]r^rircrlintstrrrvrboolrrrrrrrrrrrrr namedtuplerrrrrrrrrrrradd_fallback_dispatch_listadd_type_based_api_dispatcherrr_tf_type_based_dispatcherDispatchrrrrr rrrrrResourcer&r!r$r0r/r2r3r:r6r8r?rCrArBrErJrGrIrVrPrTrbr^r`rnrirlrzrwryrr}rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr rr rrrrr"rr!r(r%r'r0r,r3r7r6r9r?r;r=rIrGrHBoolrNrMrPrUrRrSr\r[r^rbr`rarkrirnrurqrtfloatrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr r r rrrrrrrrrr$r!r#r&r/r.r1r5r4r7r;r9r:r@r>r?rJrDFloat32rKrBrMrRrOrUrNrSr\rZr^rYr]r`rfrcrhrirorkrmrsrwrurvr~r|rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr r rrr rrrrrrrrr r$r"r#r*r'r)rGr=r;rIsCh 67179FK3IC86%%'$+,OQcey|OQbduwEGVXgixzHJZ\ln}PRbdtvFHXZi%j! ygnn1D'E S\]`bibobo]oSp }FGJLmGm}n 2I9%GHXrIstoiW^^@S6Tobkloqxq~q~l~boLUVY[|V|L}o #w~~2E(FV_`celerer`rVs4KI&IJ>4>>ZuKvwpygnnAT7Upenort{uBuBpBfCp #w~~2E(F YbcfhohuhucuYv 2I9%GHXsItupygnnAT7Uphqruw~xExEsEiFp$++MOacwzMO`bsuCETVegvxFHXZjl{}NP`brtDFVXg$h %iW^^0C&D%T]^acjcpcp^pTq%{[%ktuxzZuZk[%NG)$EF~t~~VoGpqnYsGNN?R5SnclmpryrrmdAnJjnzCDGIiDizjn%'Rr[]dhJJ3JWZJmvwz}D}K}KxKnLJX ')% &~t~~g'> ?\c\c\`c\s|~ACJCQCQ~QtR\! )C$78! RV! D1y/0 1NO b33F)GbbfbIc7>>.A$BR[\_ahanan\nRo4C "AB>4>>RiCjkl9S'..=P3Qlajknpwp}p}k}a~l!P  #w~~*= > iPSU\UcUcPcFd nwx{~TyToU hk @;I9:>4>>J];^_hygnn9L/MhU^_bdkdrdr_rUsh~GHKMcHc~dhwzhygnn)<=YWZ\c\i\iWiMj49978HZ9[\giW^^8K.Lg\efikrkxkxfx\yg///!# HiW^^(;<HIVY[b[h[hVhLiHOSHosHFIHT7)56~t~~FW7XYfYsGNN7J-Kf[dehjqjwjwew[xf_cfz~fLOf#**KM_auxKM^`qsACRTcetvDFVXhjy{LN^`prBDTVe#f:#B:UX:il:AD:[dehjqjxjxex[y:xE#CD^T^^TkElml2QlehlwzlMPl`ijmovo}o}j}`~l )C,?"@ >?i =>~t~~Nc?de3;N1O N  %% (( tIc+@&@AtyY\^e^k^kYkOltBt )&tl9978HY9Z[$5$O$O$X$X! 9S:O5O+P ^ghkmtmzmzhz^{ NQ H  %% (( JId3i&>?JyQUVYQZ\nQnGoJ@IJMOaJa@bJ)&JX3 12>4>>.3QR !/!I!I!R!R9T#Y 5M+NV_`deh`ik}`}V~OXY\^pYpOq8(*"^`gkBCBcBZ]Bpyz}@G@N@N{NqOBH +I) *>4>>*+E F __QT_cf_vADFMFTFTATwU_+-2RacjnQsQQ]`Qs|~ACJCSCS~StTQf/i-.~t~~m/LM CTWfizCDGIPIYIYDYzZD3(8(8#89SVX_XfXfSfIg4 +I) *>4>>*+E F _ #w7G7G2G(H_XabegngugubuXv_ N $i-B(BC$S\]`bibpbp]pSq$L9978HZ9[\ Ys4>>J];^_C\_nqBKLOQXQ^Q^L^B_SVX_XeXeSeIfz}NQ^aps>AcA#A3AWZAmvwz}D}J}JxJnKAF ')% &~t~~h'? @cQTcfvADFMFSFSASwT"?)TWY`YfYfTfJgDGWZgjy|>HYsGMM12HYsGMMGY=ZHmpHBEHSVHgjHT-Y+ ,^T^^K-H I Ic7==.@$AIVY[b[h[hVhLi~AQTadsv<0// H K #w}}*< =KPSKdgKuxKJMKZ7)56~t~~FX7YZygmm9K/L`csvDGVYB7 s7 37 C7 `c7 p7)56~t~~FW7XYs#Z]lo6BBSBVYBjmBAJKNPWP]P]K]A^BHI9%GHXsItuTWdgvyJSTWY`YfYfTfJg>J)C$67J)CQXQ^Q^L^B_JruJGJJX[JloJX534^T^^DT5UV3 3E)FQZ[^`g`m`m[mQnCFVYfix{@AcA#A3A_bAu~@CELERER@RvSAF534^T^^DT5UVcY\knHILNUN[N[I[\IYsGMM'9:IYsT[TaTaOaEbIuxIJMI[^IorIV;I9:>4>>J];^_Ic7==6H,IT]^acjcpcp^pTqFIY\il{~>!7 6 6 H! M #w}}2D(EMX[MloM~AMRUM^E#CD^T^^TnEopygmmAS7Thk{~LO^aB02RTfhosJCJcJbeJyBCFHOHVHVCVyWJX9978HZ9[\ggY\gkngHILNUN\N\I\]g352UWikrvYsYQTYehY|EFIKRK[K[F[|\Yv=Y;<^T^^La=bcC\_nqBKLOQXQaQaLaBbD%X qYtCy'--/G%HqPYZ^_bZceZQAqQZ[^`z[zQ{qfC "AB>4>>RiCjkId3i>V4W_himnqiruOjO`P`ijmoIjI`J8,.BbbdkoHH#H^aHt}BDKDRDRRuSHT3 12>4>>.3QR cScUXcgjc{DEHJQJXJXEX{Yc/1rQSegnrW3WSWadWxABEGNGWGWBWxXWr7)56~t~~FW7XYX[jm~GHKMTM]M]H]~^D #w~~"56 PT @-Y+ ,^T^^K-H I `ygnn1D'E``d`- 9S'*:*:%:;- UY- \1y/01OP  )C9I9I4I*J ei .)C$78.VY.`1y/0 1NO b33F)Gbfib7ygnn)<=7)CQXQ^Q^L^B_7~A7r9978HZ9[\giW^^8K.LgQZ[^`g`m`m[mQngNQgG)C1A1A,A"BGyQTV]VcVcQcGdGCFGR=Y;<^T^^La=bc3@P@P;P1QV_`celerer`rVsSV,; #w~~*= >;9SRYR_R_M_C`;B;z9978H[9\]hygnn9L/MhR[\_ahanan\nRohORhK9S'2B2B-B#CK RUW^WdWdRdHeKDGKZ=Y;<^T^^Lb=cd)CAQAQYsG,<,<'<=>[^>@534^T^^DT5UVIc7;K;K6K,Lkn*! )C$78! QT! D1y/0 1NO b33F)Gbadb& ygnn)<=& VY& N9978HZ9[\giW^^8K.Lgfig3 )C1A1A,A"B3 [^3 h=Y;<^T^^La=bc 3@P@P;P1Q kn . YsG,<,<'<=. VY. ^534^T^^DT5UV Ic7;K;K6K,L fi Ic7>>&9:)TWY`YeYeTeJf:3 12>4>>/3RS d9S'..5H+IdYbcfhohthtctYud'yg.>.>)>?'yY\^e^j^jYjOk'R7)56~t~~FX7YZ iW=M=M8M.N ^ghkmtmymyhy^z ygnn!45YsT[TaTaOaEb4 +I) *>4>>*+E F _iW^^0C&D_T]^acjcpcp^pTq_$)C)9)9$9:$)TWY`YfYfTfJg$L/i-.~t~~m/LM  38H8H3H)I YbcfhohuhucuYv 24"\]hiuvGI[]dhXcXX[XdgXqtXCFXWZXmvwz|C|J|JwJmKXt=Y;<^T^^L`=abi3icfinqiz}iKNi]`ipyz}FMMzMpNi57R_`klxyJL^`gkii[^igjitwiFIiZ]ipyz}FOOzOpPiVAy!?@PgAhi)c)fi)qt)~A)NQ)`c)s|}@BIBRBR}RsS)XYsYSYW\YqtYNQYadYz}YPYZ]_f_m_mZmPnYv/i-.~t~~l/KL cSdi~AUXfi}@PYZ]_f_m_mZmPnD-44_asvJL_artEGUWfhwyHJXZjl|~MO`brtDFVXhjy-z)& #w?O?O:O0P& ^ghkmtmzmzhz^{& HQRUW@R@HA& N$Y9-W#XYgY]YgYgiLZM$N  ygN^N^I^?_ mvwz}D}J}JxJnK W`adfOaOWP $3@P@P;P1Q$ajknpwp}p}k}a~$L%[I.Y$Z[i[_[i[ikO\P%Q!  #wO_O_J_@` pyz}@G@M@M{MqN % 3@P@P;P1Q% dmnqsztAtAoAeB% L$Y9-W#XYgY]YgYgiMZN$O   #wO_O_J_@` s|~ACJCPCP~PtQ ,33]_qtHJ]_prCESUdfuwFHVXhjz|KM^`prBDTVfhw,x(3yg>N>N9N/O3_hilnun{n{i{_|3Fn3~GHKMuHu~v3j#W),U"VWeW[WeWegIXJ#K iWM]M]H]>^ nwx{~E~K~KyKoL U} MVWZ\DWDME  +22[]orFH[]npACQSbdsuDFTVfhxzIK\^np@BRTdfu+v'M,SMfiMz}MRUMluvy{B{K{KvKlLM^"U+S!TUcUYUcUceEVF"G;bvyIL^aqz{~@G@P@P{PqQ2+22[]orFH[]npACQSbdsuDFTVfhxzIK\^np@BRTdfu+v') iW^^8K.L) ZcdgipivivdvZw) LUVY[b[h[hVhLi) |EFIKrFr|s) ENORT[TaTaOaEb) uy) T"U+S!TUcUYUcUceFVG"HvYsGNNGZ=[virsvxyFyFtFjGv[dehjqjwjwew[xvKTUXZAUAKBvT]^acjcpcp^pTqvDHv'={'='=#"($$ *11Y[mpDFY[lnAOQ`bqsBDRTdfvxGIZ\ln~@PRbds*t&,YsGNN7J-K,[dehjqjwjwew[x,Bh,\!S *Q RSaSWSaSacCTD!EuIc7>>FY4>> :Z%7ZSZXabegngugubuXvZ #w~~"56 $-Y+ ,^T^^K-H I `ygnn1D'E` 9S'*:*:%:; <1y/01OP )C9I9I4I*J N iW^^ 34AVfopsvKqKgL8 )9' ( )B C^YsGNN/B%C^Pe^u~@CEZ@Zv[^"R &3(8(8#89&F]&mvwz}TxTnU&P-Y+ ,^T^^L-I J   #w7G7G2G(H Ul }FGJLcGc}d @ ygnn!45YsNGZ=[imDMNQSaNaDbB +I) *>4>>*+E F _iW^^0C&D_IVY[iViLj_y}_NWX[]kXkNl_D ,)C)9)9$9:,)CQaLaBb,pt,KTUXZjUjKk,\/i-.~t~~m/LM 38H8H3H)IQZ[^`p[pQqAEU^_bdt_tUu"L 2ygmm!342AU2be2yBCFHOHXHXCXyY2h ')% &~t~~h'? @iW]]0B&CPdruFOPSU\UeUePeFf"; 3; s; ; TW; x # /">4>>%#893cc`c.7 7 S7 c7 Z]7 p-Y+ ,^T^^K-H I CTWfi6GiW]] 23GcGZ]GknG@CGR +I) *>4>>*+E F YsGMM/A%BVYily|LO4>>*+E F 3SVehyBCFHOHUHUCUyV<L 1ygmm!341=Q1`d1~B1[^1ENORT[TbTbObEc1f/i-.~t~~l/KL aiW]]0B&CaL`aptaIMabeaDMNQSZSaSaNaDba ygnn)<= $9978HZ9[\giW^^8K.Lg )C,?"@ <=Y;<^T^^La=bc3;N1O! )C1A1A,A"B! D=Y;<^T^^La=bc3@P@P;P1Q2;11 i %X % #w~~*= >%SVX_XgXgSgIh%rL%N;I9:>4>>J];^_hygnn9L/MhXabegngvgvbvXwhA[h3[33 i ' \ 49S'..-@#A4IVY[b[j[jVjLk4uQ4l?i =>~t~~Nd?ef)C~t~~Nd?ef)CAQAQ#SVX_XeXeSeIf#qz{~AHAPAP|PrQ#Zt#XabegAbAXB#J;I9:>4>>J];^_hygnn9L/MhXabegngtgtbtXuhAJKNPWP_P_K_A`hiChbkloqKlKbLh ' \ 19S'..-@#A1IVY[b[h[hVhLi1t}BDKDSDSSuT1]y1]fgjlHgH]I1f?i =>~t~~Nd?ef)C~t~~Nd?ef)CAQAQ!?()TWY`YhYhThJi(sv(GPQTV]VdVdQdGe(T;I9:>4>>J^;_`  #w~~:M0N YbcfhohwhwcwYx CF V_`celeses`sVt 2;11J! U30@0@+@!AUIVY[b[j[jVjLkUuxUn;I9:>4>>J^;_`  #w?O?O:O0P [dehjqjyjyey[z EH #9+"8"8J#! <38H8H3H)I<T]^acjcrcr^rTs<HQRUW^WdWdRdHe<or<|I9%GHXtIuv #wGWGWBW8XclmprysBsBnBdCW`adfmfsfsasWt~A"#T !iW^^(;<!yQTV]VeVeQeGf!pH!luvy{SvSlT!F7)56~t~~FW7XYfYsGNN7J-KfV_`celetet`tVufWfv@CE]@]v^f#T iW^^(;<YsT[TaTaOaEbmvwz}D}L}LxLnMVn~GHKMeHe~f>7)56~t~~FW7XYfYsGNN7J-KfT]^acjcpcp^pTqf}FGJLSL[L[G[}\fe}fMVWZ\tWtMuf%X *3+>!?* RUW^WdWdRdHe*pyz}@G@O@O{OqP*Ys*CLMPRlMlCm*X;I9:>4>>J^;_`  #w~~:M0N W`adfmfsfsasWt @IJMOVO^O^J^@_ hB R[\_a{\{R|  %X +30@0@+@!A+)TWY`YfYfTfJg+r{|BIBQBQ}QsR+[u+ENORTnOnEo+Z;I9:>4>>J^;_`  #w?O?O:O0P YbcfhohuhucuYv BKLOQXQ`Q`L`Ba jD T]^ac}^}T~  "R 3+>!?)TWY`YfYfTfJgpyz}@W{WqXclmpryrArAmAcBR[\_ahapap\pRq<=Y;<^T^^L`=abi #w~~:M0NiYbcfhohuhucuYvi@IJMOfJf@gir{|AHAPAP|PrQiajknpwppka@i$V (Ic7>>.A$B(YWZ\c\i\iWiMj(s|~AC\~\t](hqruw~wFwFrFhG(W`adfmfufuauWv(TAy!?@PgAhi 9S'..=P3Q \efikrkxkxfx\y CLMPRkMkCl w@ADFMFUFUAUwV fopsu|uDuDpDfE  $V ,Ic73C3C.C$D,yY\^e^k^kYkOl,u~@CE^@^v_,jstwy@yHyHtHjI,YbcfhohwhwcwYx,\Ay!?@PgAhi 9S'BRBR=R3S ^ghkmtmzmzhz^{ ENORTmOmEn yBCFHOHWHWCWyX hqruw~wFwFrFhG  iW^^(;<yQTV]VeVeQeGfvADFMFSFSASwT67)56~t~~FW7XYfYsGNN7J-KfV_`celetet`tVufFOPSU\UbUbPbFcf%3+>!?%)TWY`YhYhThJi%zCDGIPIVIVDVzW%N;I9:>4>>J^;_`  #w~~:M0N YbcfhohwhwcwYx IRSVX_XeXeSeIf '30@0@+@!A'IVY[b[j[jVjLk'|EFIKRKXKXFX|Y'R;I9:>4>>J^;_`  #w?O?O:O0P [dehjqjyjyey[z KTUXZaZgZgUgKh  N ygnn)<=iPSUjPjFkvADFMFSFSASwT_hilnun}n}i}_~NWX[]d]l]lXlNm<9978HZ9[\giW^^8K.LgU^_bdy_yUzgFOPSU\UbUbPbFcgnwx{}D}L}LxLnMg]fgjlsl{l{g{]|g"R ()C,?"@(SVXoSoIp(|EFIKRKXKXFX|Y(dmnqszsBsBnBdC(S\]`bibqbq]qSr(T=Y;<^T^^La=bc 3;N1O Xabeg~b~X KTUXZaZgZgUgKh s|}@BIBQBQ}QsR bkloqxq@q@l@bA  "R E)C1A1A,A"BE9UXZqUqKrE~GHKMTMZMZHZ~[Efopsu|uDuDpDfEEU^_bdkdsds_sUtEN=Y;<^T^^La=bc 3@P@P;P1Q ZcdgjAeA[B MVWZ\c\i\iWiMj u~BDKDSDSSuT dmnqszsBsBnBdC  !P  #w~~*= >yQTVlQlGmyBCFHOHWHWCWyXhqruw~wFwFrFhG:;I9:>4>>J];^_hygnn9L/MhV_`ce{`{V|hHQRUW^WfWfRfHghw@ADFMFUFUAUwVh!P A)C$67A@VAy}AW[AtwAJSTWY`YgYgTgJhAF3 12>4>>/3RS 3 3E)FOeEI]avyIRSVX_XfXfSfIg6.{-- v !P a)C$67a@Vay}aW[a{aY\aF3 12>4>>/3RS 3 3E)FOeEI]a}AVY@ N ygnn)<=iPSU\UbUbPbFcluvy|QwQmR]fgjlsl{l{g{]|LUVY[b[j[jVjLk<9978HZ9[\giW^^8K.LgU^_bdkdqdq_qUrg|EFIK`F`|agluvy{B{J{JvJlKg[dehjqjyjyey[zg"R ()C,?"@(SVX_XeXeSeIf(oxy|VzVpW(bkloqxq@q@l@bA(QZ[^`g`o`o[oQp(T=Y;<^T^^La=bc 3;N1O XabegngtgtbtXu HILNeIef qz{~@G@O@O{OqP `ijmovo~o~j~`  "R *)C1A1A,A"B*9UXZaZgZgUgKh*qz{~AX|XrY*dmnqszsBsBnBdC*S\]`bibqbq]qSr*X=Y;<^T^^La=bc 3@P@P;P1Q ZcdgipivivdvZw AJKNPgKgAh s|}@BIBQBQ}QsR bkloqxq@q@l@bA  ESEE#EVYEN ')% &~t~~g'> ?SPSber=