Returns metrics and plots writer.
tfma.writers.MetricsPlotsAndValidationsWriter(
output_paths: Dict[str, str],
eval_config: tfma.EvalConfig
,
add_metrics_callbacks: Optional[List[types.AddMetricsCallbackType]] = None,
metrics_key: str = constants.METRICS_KEY,
plots_key: str = constants.PLOTS_KEY,
attributions_key: str = constants.ATTRIBUTIONS_KEY,
validations_key: str = constants.VALIDATIONS_KEY,
output_file_format: str = _TFRECORD_FORMAT,
rubber_stamp: Optional[bool] = False,
stage_name: str = METRICS_PLOTS_AND_VALIDATIONS_WRITER_STAGE_NAME
) -> tfma.writers.Writer
Note, sharding will be enabled by default if a output_file_format is provided.
The files will be named -SSSSS-of-NNNNN.
where SSSSS is the shard number and NNNNN is the number of shards.
Args |
output_paths
|
Output paths keyed by output key (e.g. 'metrics', 'plots',
'validation').
|
eval_config
|
Eval config.
|
add_metrics_callbacks
|
Optional list of metric callbacks (if used).
|
metrics_key
|
Name to use for metrics key in Evaluation output.
|
plots_key
|
Name to use for plots key in Evaluation output.
|
attributions_key
|
Name to use for attributions key in Evaluation output.
|
validations_key
|
Name to use for validations key in Evaluation output.
|
output_file_format
|
File format to use when saving files. Currently
'tfrecord' and 'parquet' are supported and 'tfrecord is the default'.
If using parquet, the output metrics and plots files will contain two
columns, 'slice_key' and 'serialized_value'. The 'slice_key' column will
be a structured column matching the metrics_for_slice_pb2.SliceKey proto.
The 'serialized_value' column will contain a serialized MetricsForSlice or
PlotsForSlice proto. The validation result file will contain a single
column 'serialized_value' which will contain a single serialized
ValidationResult proto.
|
rubber_stamp
|
True if this model is being rubber stamped. When a model is
rubber stamped diff thresholds will be ignored if an associated baseline
model is not passed.
|
stage_name
|
The stage name to use when this writer is added to the Beam
pipeline.
|