1VhĽdZddlmZmZmZmZddlZddlZddlZddl Z GddejZ GddZ Gd d ZGd d ejZej"d GddZej"d GddZej"d GddZej"d GddZGddejZej"d GddZGddejZej"d GddZej"d Gdd ZGd!d"ZGd#d$eZGd%d&ZGd'd(eZGd)d*ZGd+d,eZ Gd-d.Z!Gd/d0Z"Gd1d2Z#y)3z>Experimental framework for generic TensorBoard data providers.) CollectionSequenceTupleUnionNc@eZdZdZddZddZejddZej ddddZ ej ddddd Z ej dddd Z dddd Z ddddd Z dddd Z dddddZddZddddZ ddddddZy) DataProvidera}Interface for reading TensorBoard scalar, tensor, and blob data. These APIs are under development and subject to change. For instance, providers may be asked to implement more filtering mechanisms, such as downsampling strategies or domain restriction by step or wall time. The data provider interface specifies three *data classes*: scalars, tensors, and blob sequences. All data is stored in *time series* for one of these data classes. A time series is identified by run name and tag name (each a non-empty text string), as well as an experiment ID and plugin name (see below). Points in a time series are uniquely indexed by *step*, an arbitrary non-negative integer. Each point in a time series also has an associated wall time, plus its actual value, which is drawn from the corresponding data class. Each point in a scalar time series contains a single scalar value, as a 64-bit floating point number. Scalars are "privileged" rather than being subsumed under tensors because there are useful operations on scalars that don't make sense in the general tensor case: e.g., "list all scalar time series with tag name `accuracy` whose exponentially weighted moving average is at least 0.999". Each point in a tensor time series contains a tensor of arbitrary dtype (including byte strings and text strings) and shape (including rank-0 tensors, a.k.a. scalars). Each tensor is expected to be "reasonably small" to accommodate common database cell size limits. For instance, a histogram with a bounded number of buckets (say, 30) occupies about 500 bytes, and a PR curve with a bounded number of thresholds (say, 201) occupies about 5000 bytes. These are both well within typical database tolerances (Google Cloud Spanner: 10 MiB; MySQL: 64 KiB), and would be appropriate to store as tensors. By contrast, image, audio, or model graph data may easily be multiple megabytes in size, and so should be stored as blobs instead. The tensors at each step in a time series need not have the same dtype or shape. Each point in a blob sequence time series contains an ordered sequence of zero or more blobs, which are arbitrary data with no tensor structure. These might represent PNG-encoded image data, protobuf wire encodings of TensorFlow graphs, or PLY-format 3D mesh data, for some examples. This data class provides blob *sequences* rather than just blobs because it's common to want to take multiple homogeneous samples of a given time series: say, "show me the bounding box classifications for 3 random inputs from this batch". A single blob can of course be represented as a blob sequence that always has exactly one element. When reading time series, *downsampling* refers to selecting a subset of the points in each time series. Downsampling only occurs across the step axis (rather than, e.g., the blobs in a single blob sequence datum), and occurs individually within each time series. When downsampling, the latest datum should always be included in the sample, so that clients have a view of metrics that is maximally up to date. Implementations may choose to force the first (oldest) datum to be included in each sample as well, but this is not required; clients should not make assumptions either way. The remainder of the points in the sample should be selected uniformly at random from available points. Downsampling should be deterministic within a time series. It is also useful for the downsampling behavior to depend only on the set of step values within a time series, such that two "parallel" time series with data at exactly the same steps also retain the same steps after downsampling. Every time series belongs to a specific experiment and is owned by a specific plugin. (Thus, the "primary key" for a time series has four components: experiment, plugin, run, tag.) The experiment ID is an arbitrary URL-safe non-empty text string, whose interpretation is at the discretion of the data provider. As a special case, the empty string as an experiment ID denotes that no experiment was given. Data providers may or may not fully support an empty experiment ID. The plugin name should correspond to the `plugin_data.plugin_name` field of the `SummaryMetadata` proto passed to `tf.summary.write`. Additionally, the data provider interface specifies one *hyperparameter* class, which is metadata about the parameters used to generate the data for one or more runs within one or more experiments. Each hyperparameter has a value type -- one of string, bool, and float. Each one also has a domain, which describes the set of known values for that hyperparameter across the given set of experiments. There is a corresponding *hyperparameter value* class, which describes an actual value of a hyperparameter that was logged during experiment execution. Each run within an experiment may specify its own value for a hyperparameter. Runs that were logically executed together with the same set of hyperparameter values form a hyperparameter `session`. Sessions that include the same hyperparameter values can be grouped together in a hyperparameter `session group`. Often a session group will contain only a single session. However, in some scenarios, the same hyperparameters will be used to execute multiple jobs with the idea to aggregate the metrics across those jobs and analyze non-deterministic factors. In that case, a session group will contain multiple sessions. The result will group runs by hyperparameter session group and provide one set of hyperparameter values for each group. All methods on this class take a `RequestContext` parameter as the first positional argument. This argument is temporarily optional to facilitate migration, but will be required in the future. Unless otherwise noted, any methods on this class may raise errors defined in `tensorboard.errors`, like `tensorboard.errors.NotFoundError`. If not implemented, optional methods may return `None`. NctS)aRetrieve metadata of a given experiment. The metadata may include fields such as name and description of the experiment, as well as a timestamp for the experiment. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of the experiment in question. Returns: An `ExperimentMetadata` object containing metadata about the experiment. )ExperimentMetadataselfctx experiment_ids I/home/dcms/DCMS/lib/python3.12/site-packages/tensorboard/data/provider.pyexperiment_metadataz DataProvider.experiment_metadatas "##cy)aList all plugins that own data in a given experiment. This should be the set of all plugin names `p` such that calling `list_scalars`, `list_tensors`, or `list_blob_sequences` for the given `experiment_id` and plugin name `p` gives a non-empty result. This operation is optional, but may later become required. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. Returns: A collection of strings representing plugin names, or `None` if this operation is not supported by this data provider. Nr s r list_pluginszDataProvider.list_pluginss$rcy)a>List all runs within an experiment. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. Returns: A collection of `Run` values. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nrr s r list_runszDataProvider.list_runs r)run_tag_filtercy)aList metadata about scalar time series. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. plugin_name: String name of the TensorBoard plugin that created the data to be queried. Required. run_tag_filter: Optional `RunTagFilter` value. If omitted, all runs and tags will be included. The result will only contain keys for run-tag combinations that actually exist, which may not include all entries in the `run_tag_filter`. Returns: A nested map `d` such that `d[run][tag]` is a `ScalarTimeSeries` value. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nrr r r plugin_namers r list_scalarszDataProvider.list_scalarss2 r) downsamplercy)aRead values from scalar time series. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. plugin_name: String name of the TensorBoard plugin that created the data to be queried. Required. downsample: Integer number of steps to which to downsample the results (e.g., `1000`). The most recent datum (last scalar) should always be included. See `DataProvider` class docstring for details about this parameter. Required. run_tag_filter: Optional `RunTagFilter` value. If provided, a time series will only be included in the result if its run and tag both pass this filter. If `None`, all time series will be included. The result will only contain keys for run-tag combinations that actually exist, which may not include all entries in the `run_tag_filter`. Returns: A nested map `d` such that `d[run][tag]` is a list of `ScalarDatum` values sorted by step. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nrr r rrrrs r read_scalarszDataProvider.read_scalarssJ rcy)aRead the most recent values from scalar time series. The most recent scalar value for each tag under each run is retrieved from the latest event (at the latest timestamp). Note that this is different from the sorting used in `read_scalars`, which is by step. This was an accidental misalignment that would need considerable effort to change across our implementations, so we're leaving it as is for now. In most cases this should not matter, but if the same log dir is used for multiple runs, this might not match the last data point returned by the `read_scalars`. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. plugin_name: String name of the TensorBoard plugin that created the data to be queried. Required. run_tag_filter: Optional `RunTagFilter` value. If provided, a datum series will only be included in the result if its run and tag both pass this filter. If `None`, all time series will be included. The result will only contain keys for run-tag combinations that actually exist, which may not include all entries in the `run_tag_filter`. Returns: A nested map `d` such that `d[run][tag]` is a `ScalarDatum` representing the latest scalar in the time series. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nrrs rread_last_scalarszDataProvider.read_last_scalarssR rcy)aList metadata about tensor time series. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. plugin_name: String name of the TensorBoard plugin that created the data to be queried. Required. run_tag_filter: Optional `RunTagFilter` value. If omitted, all runs and tags will be included. The result will only contain keys for run-tag combinations that actually exist, which may not include all entries in the `run_tag_filter`. Returns: A nested map `d` such that `d[run][tag]` is a `TensorTimeSeries` value. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nrrs r list_tensorszDataProvider.list_tensors&s0 rcy)a]Read values from tensor time series. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. plugin_name: String name of the TensorBoard plugin that created the data to be queried. Required. downsample: Integer number of steps to which to downsample the results (e.g., `1000`). See `DataProvider` class docstring for details about this parameter. Required. run_tag_filter: Optional `RunTagFilter` value. If provided, a time series will only be included in the result if its run and tag both pass this filter. If `None`, all time series will be included. The result will only contain keys for run-tag combinations that actually exist, which may not include all entries in the `run_tag_filter`. Returns: A nested map `d` such that `d[run][tag]` is a list of `TensorDatum` values sorted by step. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nrrs r read_tensorszDataProvider.read_tensors@sF rcy)aList metadata about blob sequence time series. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. plugin_name: String name of the TensorBoard plugin that created the data to be queried. Required. run_tag_filter: Optional `RunTagFilter` value. If omitted, all runs and tags will be included. The result will only contain keys for run-tag combinations that actually exist, which may not include all entries in the `run_tag_filter`. Returns: A nested map `d` such that `d[run][tag]` is a `BlobSequenceTimeSeries` value. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nrrs rlist_blob_sequencesz DataProvider.list_blob_sequenceses, rcy)a]Read values from blob sequence time series. Args: ctx: A TensorBoard `RequestContext` value. experiment_id: ID of enclosing experiment. plugin_name: String name of the TensorBoard plugin that created the data to be queried. Required. downsample: Integer number of steps to which to downsample the results (e.g., `1000`). See `DataProvider` class docstring for details about this parameter. Required. run_tag_filter: Optional `RunTagFilter` value. If provided, a time series will only be included in the result if its run and tag both pass this filter. If `None`, all time series will be included. The result will only contain keys for run-tag combinations that actually exist, which may not include all entries in the `run_tag_filter`. Returns: A nested map `d` such that `d[run][tag]` is a list of `BlobSequenceDatum` values sorted by step. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nrrs rread_blob_sequencesz DataProvider.read_blob_sequences}s@ rcy)aoRead data for a single blob. Args: ctx: A TensorBoard `RequestContext` value. blob_key: A key identifying the desired blob, as provided by `read_blob_sequences(...)`. Returns: Raw binary data as `bytes`. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. Nr)r r blob_keys r read_blobzDataProvider.read_blobrr)limitctggS)aCList hyperparameters metadata. Args: ctx: A TensorBoard `RequestContext` value. experiment_ids: A Collection[string] of IDs of the enclosing experiments. limit: Optional number of hyperparameter metadata to include in the result. If unset or zero, all metadata will be included. Returns: A ListHyperparametersResult describing the hyperparameter-related metadata for the experiments. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. )hyperparameterssession_groups)ListHyperparametersResult)r r experiment_idsr.s rlist_hyperparametersz!DataProvider.list_hyperparameterss")BOOr)filterssorthparams_to_includecgS)aRead hyperparameter values. Args: ctx: A TensorBoard `RequestContext` value. experiment_ids: A Collection[string] of IDs of the enclosing experiments. filters: A Collection[HyperparameterFilter] that constrain the returned session groups based on hyperparameter value. sort: A Sequence[HyperparameterSort] that specify how the results should be sorted. hparams_to_include: An optional Collection[str] of the full names of hyperparameters to include in the results. This collection will be augmented to include all the hyperparameters specified in `filters` and `sort`. If None, all hyperparameters will be returned. Returns: A Sequence[HyperparameterSessionGroup] describing the groups and their hyperparameter values. Raises: tensorboard.errors.PublicError: See `DataProvider` class docstring. r)r r r3r5r6r7s rread_hyperparametersz!DataProvider.read_hyperparameterss > rN)__name__ __module__ __qualname____doc__rrabcabstractmethodrrr r"r$r&r(r*r-r4r9rrrrrs*hT$ (       FJ  4  $ $ $ L  (  ( ( V FJ 8 # # L FJ 4     D PdP*   rr) metaclassczeZdZdZddddddZedZedZedZed Z d Z d Z d Z d Z y)r aIMetadata about an experiment. All fields have default values: i.e., they will always be present on the object, but may be omitted in a constructor call. Attributes: data_location: A human-readable description of the data source, such as a path to a directory on disk. experiment_name: A user-facing name for the experiment (as a `str`). experiment_description: A user-facing description for the experiment (as a `str`). creation_time: A timestamp for the creation of the experiment, as `float` seconds since the epoch. r) data_locationexperiment_nameexperiment_description creation_timec<||_||_||_||_yr:_data_location_experiment_name_experiment_description_creation_time)r rDrErFrGs r__init__zExperimentMetadata.__init__s$, /'=$+rc|jSr:)rJr s rrDz ExperimentMetadata.data_location"""rc|jSr:)rKrPs rrEz"ExperimentMetadata.experiment_names$$$rc|jSr:)rLrPs rrFz)ExperimentMetadata.experiment_description s+++rc|jSr:)rMrPs rrGz ExperimentMetadata.creation_time rQrc^|j|j|j|jfS)z#Helper for `__eq__` and `__hash__`.rIrPs r _as_tuplezExperimentMetadata._as_tuples2     ! !  ( (      rcft|tsy|j|jk(S)NF) isinstancer rVr others r__eq__zExperimentMetadata.__eq__s)%!34~~5??#444rc4t|jSr:)hashrVrPs r__hash__zExperimentMetadata.__hash__sDNN$%%rcddjd|jd|jd|jd|jfzS)NzExperimentMetadata(%s), zdata_location=zexperiment_name=zexperiment_description=zcreation_time=)joinrDrKrLrMrPs r__repr__zExperimentMetadata.__repr__"sO'$))&*&8&8:(,(=(=?/3/K/KM&*&9&9;  +   rN)r;r<r=r>rNpropertyrDrErFrGrVr[r^rbrrrr r s~ $! ,##%%,,## 5 & rr c\eZdZdZdZdZedZedZedZ dZ dZ d Z y ) RunaQMetadata about a run. Attributes: run_id: A unique opaque string identifier for this run. run_name: A user-facing name for this run (as a `str`). start_time: The wall time of the earliest recorded event in this run, as `float` seconds since epoch, or `None` if this run has no recorded events. _run_id _run_name _start_timec.||_||_||_yr:rf)r run_idrun_name start_times rrNz Run.__init__:s !%rc|jSr:)rgrPs rrkz Run.run_id? ||rc|jSr:)rhrPs rrlz Run.run_nameC ~~rc|jSr:)rirPs rrmzRun.start_timeGrct|tsy|j|jk7ry|j|jk7ry|j|jk7ryyNFT)rXrergrhrirYs rr[z Run.__eq__KsO%% <<5== ( >>U__ ,   u00 0rcZt|j|j|jfSr:)r]rgrhrirPs rr^z Run.__hash__Vs"T\\4>>43C3CDEErc~ddjd|jd|jd|jfzS)NzRun(%s)r`zrun_id=z run_name=z start_time=)rargrhrirPs rrbz Run.__repr__Ys?499#||-!%1#'#3#35    rN) r;r<r=r> __slots__rNrcrkrlrmr[r^rbrrrrere-s_8I&    F rrec eZdZdZdZdZdZdZy)HyperparameterDomainTypezHDescribes how to represent the set of known values for a hyperparameter.intervaldiscrete_floatdiscrete_string discrete_boolN)r;r<r=r>INTERVALDISCRETE_FLOATDISCRETE_STRING DISCRETE_BOOLrrrrzrzcsRH%N'O#MrrzT)frozenceZdZUdZeed<eed<dZeedfed<dZ ee e e fe e e ee e dfed<dZe ed<y) Hyperparametera#Metadata about a hyperparameter. Attributes: hyperparameter_name: A string identifier for the hyperparameter that should be unique in any result set of Hyperparameter objects. hyperparameter_display_name: A displayable name for the hyperparameter. Unlike hyperparameter_name, there is no uniqueness constraint. domain_type: A HyperparameterDomainType describing how we represent the set of known values in the `domain` attribute. domain: A representation of the set of known values for the hyperparameter. If domain_type is INTERVAL, a Tuple[float, float] describing the range of numeric values. If domain_type is DISCRETE_FLOAT, a Collection[float] describing the finite set of numeric values. If domain_type is DISCRETE_STRING, a Collection[string] describing the finite set of string values. If domain_type is DISCRETE_BOOL, a Collection[bool] describing the finite set of bool values. differs: Describes whether there are two or more known values for the hyperparameter for the set of experiments specified in the list_hyperparameters() request. Hyperparameters for which this is true are made more prominent or easier to discover in the UI. hyperparameter_namehyperparameter_display_nameN domain_typedomainFdiffers)r;r<r=r>str__annotations__rrrzrrfloatrboolrrrrrrpsw6!$$9=K/56=  E eUl534    GTrrcPeZdZUdZeed<dZeedfed<dZ ee ee dfed<y)HyperparameterValuea|A hyperparameter value. Attributes: hyperparameter_name: A string identifier for the hyperparameters. It corresponds to the hyperparameter_name field in the Hyperparameter class. domain_type: A HyperparameterDomainType describing how we represent the set of known values in the `domain` attribute. value: The value of the hyperparameter. If domain_type is INTERVAL or DISCRETE_FLOAT, value is a float. If domain_type is DISCRETE_STRING, value is a str. If domain_type is DISCRETE_BOOL, value is a bool. If domain_type is unknown (None), value is None. rNrvalue) r;r<r=r>rrrrrzrrrrrrrrs< 9=K/56=+/E5T4' (/rrc&eZdZUdZeed<eed<y)HyperparameterSessionRunzA single run in a HyperparameterSessionGroup. Attributes: experiment_id: The id of the experiment to which the run belongs. run: The name of the run. rrunN)r;r<r=r>rrrrrrrs Hrrc<eZdZUdZeed<eeed<eeed<y)HyperparameterSessionGroupaA group of sessions logically executed together with the same hparam values. A `session` generally represents a particular execution of a job with a given set of hyperparameter values. A session may contain multiple related runs executed together to train and/or validate a model. Often a `session group` will contain only a single session. However, in some scenarios, the same hyperparameters will be used to execute multiple jobs with the idea to aggregate the metrics across those jobs and analyze non-deterministic factors. In that case, a session group will contain multiple sessions. Attributes: root: A descriptor of the common ancestor of all sessions in this group. In the case where the group contains all runs in the experiment, this would just be a HyperparameterSessionRun with the experiment_id property set to the experiment's id but run property set to empty. In the case where the group contains a subset of runs in the experiment, this would be a HyperparameterSessionRun with the experiment_id property set and the run property set to the largest common prefix for runs. The root might correspond to a session within the group but it is not necessary. sessions: A sequence of all sessions in this group. hyperparameter_values: A collection of all hyperparameter values in this group. rootsessionshyperparameter_valuesN) r;r<r=r>rrrrrrrrrrs'> #"/00%&9::rrceZdZdZdZdZdZy)HyperparameterFilterTypez)Describes how to represent filter values.regexr{discreteN)r;r<r=r>REGEXrDISCRETErrrrrs3 EHHrrcXeZdZUdZeed<eed<eeee e fe ee ee ffed<y)HyperparameterFilteraA constraint based on hyperparameter value. Attributes: hyperparameter_name: A string identifier for the hyperparameter to use for the filter. It corresponds to the hyperparameter_name field in the Hyperparameter class. filter_type: A HyperparameterFilterType describing how we represent the filter values in the 'filter' attribute. filter: A representation of the set of the filter values. If filter_type is REGEX, a str containing the regular expression. If filter_type is INTERVAL, a Tuple[float, float] describing the min and max values of the filter interval. If filter_type is DISCRETE a Collection[float|str|bool] describing the finite set of filter values. r filter_typefilterN) r;r<r=r>rrrrrrrrrrrrrsJ"))   eUl5T)*+ - rrceZdZdZdZdZy)HyperparameterSortDirectionz*Describes which direction to sort a value. ascending descendingN)r;r<r=r> ASCENDING DESCENDINGrrrrr s4IJrrc&eZdZUdZeed<eed<y)HyperparameterSorta*A sort criterium based on hyperparameter value. Attributes: hyperparameter_name: A string identifier for the hyperparameter to use for the sort. It corresponds to the hyperparameter_name field in the Hyperparameter class. sort_direction: The direction to sort. rsort_directionN)r;r<r=r>rrrrrrrrs//rrc2eZdZUdZeeed<eeed<y)r2aThe result from calling list_hyperparameters(). Attributes: hyperparameters: The hyperparameteres belonging to the experiments in the request. session_groups: The session groups present in the experiments in the request. r0r1N)r;r<r=r>rrrrrrrr2r2#s  //9::rr2ceZdZdZdZdddZedZedZedZ ed Z ed Z ed Z y) _TimeSerieszMetadata about time series data for a particular run and tag. Superclass of `ScalarTimeSeries`, `TensorTimeSeries`, and `BlobSequenceTimeSeries`.  _max_step_max_wall_time_plugin_content _description _display_name _last_valueN) last_valuecX||_||_||_||_||_||_yr:r)r max_step max_wall_timeplugin_content description display_namers rrNz_TimeSeries.__init__Bs3"+-')%rc|jSr:)rrPs rrz_TimeSeries.max_stepSrqrc|jSr:)rrPs rrz_TimeSeries.max_wall_timeWrQrc|jSr:)rrPs rrz_TimeSeries.plugin_content[s###rc|jSr:)rrPs rrz_TimeSeries.description_s   rc|jSr:)rrPs rrz_TimeSeries.display_namecs!!!rc|jSr:)rrPs rrz_TimeSeries.last_valuegrsr) r;r<r=r>rxrNrcrrrrrrrrrrr2s I"&"##$$!!""  rrc"eZdZdZdZdZdZy)ScalarTimeSeriesa Metadata about a scalar time series for a particular run and tag. Attributes: max_step: The largest step value of any datum in this scalar time series; a nonnegative integer. max_wall_time: The largest wall time of any datum in this time series, as `float` seconds since epoch. plugin_content: A bytestring of arbitrary plugin-specific metadata for this time series, as provided to `tf.summary.write` in the `plugin_data.content` field of the `metadata` argument. description: An optional long-form Markdown description, as a `str` that is empty if no description was specified. display_name: An optional long-form Markdown description, as a `str` that is empty if no description was specified. Deprecated; may be removed soon. last_value: An optional value for the latest scalar in the time series, corresponding to the scalar at `max_step`. Note that this field might NOT be populated by all data provider implementations. c^t|tsy|j|jk7ry|j|jk7ry|j|jk7ry|j |j k7ry|j |j k7ry|j|jk7ryyru)rXrrrrrrrrYs rr[zScalarTimeSeries.__eq__s%!12 >>U__ ,   %"6"6 6   5#8#8 8    2 2 2   !4!4 4   u00 0rct|j|j|j|j|j |j fSr:)r]rrrrrrrPs rr^zScalarTimeSeries.__hash__sH##$$!!""     rc ddjd|jd|jd|jd|jd|j d|j fzS) NzScalarTimeSeries(%s)r` max_step=max_wall_time=plugin_content= description= display_name=z last_value=)rarrrrrrrPs rrbzScalarTimeSeries.__repr__se% !%1&*&9&9;'+';';=$($5$57%)%7%79#'#3#35   )  rNr;r<r=r>r[r^rbrrrrrls&"    rrc\eZdZdZdZdZedZedZedZ dZ dZ d Z y ) ScalarDatumavA single datum in a scalar time series for a run and tag. Attributes: step: The global step at which this datum occurred; an integer. This is a unique key among data of this time series. wall_time: The real-world time at which this datum occurred, as `float` seconds since epoch. value: The scalar value for this datum; a `float`. _step _wall_time_valuec.||_||_||_yr:r)r step wall_timers rrNzScalarDatum.__init__ # rc|jSr:rrPs rrzScalarDatum.step zzrc|jSr:rrPs rrzScalarDatum.wall_time rc|jSr:)rrPs rrzScalarDatum.value {{rct|tsy|j|jk7ry|j|jk7ry|j|jk7ryyru)rXrrrrrYs rr[zScalarDatum.__eq__sM%- :: $ ??e.. . ;;%,, &rcZt|j|j|jfSr:)r]rrrrPs rr^zScalarDatum.__hash__s TZZ$++>??rc~ddjd|jd|jd|jfzS)NzScalarDatum(%s)r`step= wall_time=zvalue=)rarrrrPs rrbzScalarDatum.__repr__= 499!ZZ)"&//3"kk+ $   rN) r;r<r=r>rxrNrcrrrr[r^rbrrrrrs_2I  @ rrc"eZdZdZdZdZdZy)TensorTimeSeriesa/Metadata about a tensor time series for a particular run and tag. Attributes: max_step: The largest step value of any datum in this tensor time series; a nonnegative integer. max_wall_time: The largest wall time of any datum in this time series, as `float` seconds since epoch. plugin_content: A bytestring of arbitrary plugin-specific metadata for this time series, as provided to `tf.summary.write` in the `plugin_data.content` field of the `metadata` argument. description: An optional long-form Markdown description, as a `str` that is empty if no description was specified. display_name: An optional long-form Markdown description, as a `str` that is empty if no description was specified. Deprecated; may be removed soon. c*t|tsy|j|jk7ry|j|jk7ry|j|jk7ry|j |j k7ry|j |j k7ryyru)rXrrrrrrrYs rr[zTensorTimeSeries.__eq__s%!12 >>U__ ,   %"6"6 6   5#8#8 8    2 2 2   !4!4 4rct|j|j|j|j|j fSr:)r]rrrrrrPs rr^zTensorTimeSeries.__hash__s?##$$!!""    rc ddjd|jd|jd|jd|jd|j fzS)NzTensorTimeSeries(%s)r`rrrrr)rarrrrrrPs rrbzTensorTimeSeries.__repr__ sY% !%1&*&9&9;'+';';=$($5$57%)%7%79  )   rNrrrrrrs     rrcZeZdZdZdZdZedZedZedZ dZ dZ d Z y) TensorDatumaA single datum in a tensor time series for a run and tag. Attributes: step: The global step at which this datum occurred; an integer. This is a unique key among data of this time series. wall_time: The real-world time at which this datum occurred, as `float` seconds since epoch. numpy: The `numpy.ndarray` value with the tensor contents of this datum. rr_numpyc.||_||_||_yr:r)r rrnumpys rrNzTensorDatum.__init__%rrc|jSr:rrPs rrzTensorDatum.step*rrc|jSr:rrPs rrzTensorDatum.wall_time.rrc|jSr:)rrPs rrzTensorDatum.numpy2rrct|tsy|j|jk7ry|j|jk7ryt j |j |j syyru)rXrrrnp array_equalrrYs rr[zTensorDatum.__eq__6sS%- :: $ ??e.. .~~dkk5<<8rNc~ddjd|jd|jd|jfzS)NzTensorDatum(%s)r`rrznumpy=)rarrrrPs rrbzTensorDatum.__repr__Drr) r;r<r=r>rxrNrcrrrr[r^rbrrrrrs` 2I  H rrcHeZdZdZdZfdZedZdZdZ dZ xZ S)BlobSequenceTimeSeriesaMetadata about a blob sequence time series for a particular run and tag. Attributes: max_step: The largest step value of any datum in this scalar time series; a nonnegative integer. max_wall_time: The largest wall time of any datum in this time series, as `float` seconds since epoch. max_length: The largest length (number of blobs) of any datum in this scalar time series, or `None` if this time series is empty. plugin_content: A bytestring of arbitrary plugin-specific metadata for this time series, as provided to `tf.summary.write` in the `plugin_data.content` field of the `metadata` argument. description: An optional long-form Markdown description, as a `str` that is empty if no description was specified. display_name: An optional long-form Markdown description, as a `str` that is empty if no description was specified. Deprecated; may be removed soon.  _max_lengthc<t||||||||_y)N)rrrrr)superrNr)r rr max_lengthrrr __class__s rrNzBlobSequenceTimeSeries.__init__cs0 ')#%  &rc|jSr:rrPs rrz!BlobSequenceTimeSeries.max_lengthvrsrc^t|tsy|j|jk7ry|j|jk7ry|j|jk7ry|j |j k7ry|j |j k7ry|j|jk7ryyru)rXrrrrrrrrYs rr[zBlobSequenceTimeSeries.__eq__zs%!78 >>U__ ,   %"6"6 6   u00 0   5#8#8 8    2 2 2   !4!4 4rct|j|j|j|j|j |j fSr:)r]rrrrrrrPs rr^zBlobSequenceTimeSeries.__hash__sH##  $$!!""    rc ddjd|jd|jd|jd|jd|j d|j fzS) NzBlobSequenceTimeSeries(%s)r`rrz max_length=rrr)rarrrrrrrPs rrbzBlobSequenceTimeSeries.__repr__se+dii!%1&*&9&9;#'#3#35'+';';=$($5$57%)%7%79   /  r) r;r<r=r>rxrNrcrr[r^rb __classcell__)rs@rrrNs7$!I&&  "    rrcNeZdZdZdZd dZedZedZdZ dZ d Z y) BlobReferenceaAA reference to a blob. Attributes: blob_key: A string containing a key uniquely identifying a blob, which may be dereferenced via `provider.read_blob(blob_key)`. These keys must be constructed such that they can be included directly in a URL, with no further encoding. Concretely, this means that they consist exclusively of "unreserved characters" per RFC 3986, namely [a-zA-Z0-9._~-]. These keys are case-sensitive; it may be wise for implementations to normalize case to reduce confusion. The empty string is not a valid key. Blob keys must not contain information that should be kept secret. Privacy-sensitive applications should use random keys (e.g. UUIDs), or encrypt keys containing secret fields. url: (optional) A string containing a URL from which the blob data may be fetched directly, bypassing the data provider. URLs may be a vector for data leaks (e.g. via browser history, web proxies, etc.), so these URLs should not expose secret information. )_url _blob_keyNc ||_||_yr:)r r)r r,urls rrNzBlobReference.__init__s! rc|jS)a;Provide a key uniquely identifying a blob. Callers should consider these keys to be opaque-- i.e., to have no intrinsic meaning. Some data providers may use random IDs; but others may encode information into the key, in which case callers must make no attempt to decode it. )r rPs rr,zBlobReference.blob_keys~~rc|jS)aBProvide the direct-access URL for this blob, if available. Note that this method is *not* expected to construct a URL to the data-loading endpoint provided by TensorBoard. If this method returns None, then the caller should proceed to use `blob_key()` to build the URL, as needed. )rrPs rr zBlobReference.urlsyyrct|tsy|j|jk7ry|j|jk7ryyru)rXrr rrYs rr[zBlobReference.__eq__s8%/ >>U__ , 99 "rcDt|j|jfSr:)r]r rrPs rr^zBlobReference.__hash__sT^^TYY/00rcbddjd|jd|jfzS)NzBlobReference(%s)r`z blob_key=zurl=)rar rrPs rrbzBlobReference.__repr__s0"TYY"nn .DII0G H&   rr:) r;r<r=r>rxrNrcr,r r[r^rbrrrrrsJ,&I1 rrc\eZdZdZdZdZedZedZedZ dZ dZ d Z y ) BlobSequenceDatumaA single datum in a blob sequence time series for a run and tag. Attributes: step: The global step at which this datum occurred; an integer. This is a unique key among data of this time series. wall_time: The real-world time at which this datum occurred, as `float` seconds since epoch. values: A tuple of `BlobReference` objects, providing access to elements of this sequence. rr_valuesc.||_||_||_yr:r)r rrvaluess rrNzBlobSequenceDatum.__init__s # rc|jSr:rrPs rrzBlobSequenceDatum.steprrc|jSr:rrPs rrzBlobSequenceDatum.wall_timerrc|jSr:)rrPs rrzBlobSequenceDatum.valuesrorct|tsy|j|jk7ry|j|jk7ry|j|jk7ryyru)rXrrrrrYs rr[zBlobSequenceDatum.__eq__sN%!23 :: $ ??e.. . <<5== (rcZt|j|j|jfSr:)r]rrrrPs rr^zBlobSequenceDatum.__hash__s TZZ$,,?@@rc~ddjd|jd|jd|jfzS)NzBlobSequenceDatum(%s)r`rrzvalues=)rarrrrPs rrbzBlobSequenceDatum.__repr__s=&!ZZ)"&//3#||- *   rN) r;r<r=r>rxrNrcrrrr[r^rbrrrrrs_ 3I  A rrcDeZdZdZddZdZedZedZdZ y) RunTagFilterz"Filters data by run and tag names.Nc`|jd||_|jd||_y)aConstruct a `RunTagFilter`. A time series passes this filter if both its run *and* its tag are included in the corresponding whitelists. Order and multiplicity are ignored; `runs` and `tags` are treated as sets. Args: runs: Collection of run names, as strings, or `None` to admit all runs. tags: Collection of tag names, as strings, or `None` to admit all tags. runstagsN)_parse_optional_string_set_runs_tags)r r r!s rrNzRunTagFilter.__init__#s,44VTB 44VTB rc|yt|trt|dt|d|t |}|D].}t|trt|dt|d||S)Nz0: expected `None` or collection of strings; got z: z=: expected `None` or collection of strings; got item of type )rXr TypeErrortype frozenset)r nameritems rr"z'RunTagFilter._parse_optional_string_set5s} = eS !ee- %  DdC(15tDz4I   rc|jSr:)r#rPs rr zRunTagFilter.runsHrrc|jSr:)r$rPs rr!zRunTagFilter.tagsLrrcbddjd|jd|jfzS)NzRunTagFilter(%s)r`zruns=ztags=)rar#r$rPs rrbzRunTagFilter.__repr__Ps3!DII!ZZ)!ZZ) %   r)NN) r;r<r=r>rNr"rcr r!rbrrrrr s=,C$& rr)$r>typingrrrrr? dataclassesenumrrABCMetarr reEnumrz dataclassrrrrrrrrr2rrrrrrrrrrrrr4sE65  GS[[GTF F R3 3 l $tyy $d#&&$&Rd#00$0,d#   $  d#";";$";Jtyyd#$6$))d# 0 0$ 0d# ; ;$ ;7 7 t; {; |3 3 l4 {4 n4 4 nS [S lB B J4 4 n6 6 r