Options for generating statistics.
tfdv.StatsOptions(
generators: Optional[List[stats_generator.StatsGenerator]] = None,
schema: Optional[schema_pb2.Schema] = None,
label_feature: Optional[types.FeatureName] = None,
weight_feature: Optional[types.FeatureName] = None,
slice_functions: Optional[List[types.SliceFunction]] = None,
sample_rate: Optional[float] = None,
num_top_values: int = 20,
frequency_threshold: int = 1,
weighted_frequency_threshold: float = 1.0,
num_rank_histogram_buckets: int = 1000,
num_values_histogram_buckets: int = 10,
num_histogram_buckets: int = 10,
num_quantiles_histogram_buckets: int = 10,
epsilon: float = 0.01,
infer_type_from_schema: bool = False,
desired_batch_size: Optional[int] = None,
enable_semantic_domain_stats: bool = False,
semantic_domain_stats_sample_rate: Optional[float] = None,
per_feature_weight_override: Optional[Dict[types.FeaturePath, types.FeatureName]] = None,
vocab_paths: Optional[Dict[types.VocabName, types.VocabPath]] = None,
add_default_generators: bool = True,
feature_allowlist: Optional[Union[List[types.FeatureName], List[types.FeaturePath]]] = None,
experimental_use_sketch_based_topk_uniques: Optional[bool] = None,
use_sketch_based_topk_uniques: Optional[bool] = None,
experimental_slice_functions: Optional[List[types.SliceFunction]] = None,
experimental_slice_sqls: Optional[List[Text]] = None,
experimental_result_partitions: int = 1,
experimental_num_feature_partitions: int = 1,
slicing_config: Optional[slicing_spec_pb2.SlicingConfig] = None,
experimental_filter_read_paths: bool = False,
per_feature_stats_config: Optional[types.PerFeatureStatsConfig] = None
)
Args |
|---|
generators
|
An optional list of statistics generators. A statistics
generator must extend either CombinerStatsGenerator or
TransformStatsGenerator.
|
schema
|
An optional tensorflow_metadata Schema proto. Currently we use the
schema to infer categorical and bytes features.
|
label_feature
|
An optional feature name which represents the label.
|
weight_feature
|
An optional feature name whose numeric value represents
the weight of an example.
|
slice_functions
|
DEPRECATED. Use experimental_slice_functions.
|
sample_rate
|
An optional sampling rate. If specified, statistics is
computed over the sample.
|
num_top_values
|
An optional number of most frequent feature values to keep
for string features.
|
frequency_threshold
|
An optional minimum number of examples the most
frequent values must be present in.
|
weighted_frequency_threshold
|
An optional minimum weighted number of
examples the most frequent weighted values must be present in. This
option is only relevant when a weight_feature is specified.
|
num_rank_histogram_buckets
|
An optional number of buckets in the rank
histogram for string features.
|
num_values_histogram_buckets
|
An optional number of buckets in a quantiles
histogram for the number of values per Feature, which is stored in
CommonStatistics.num_values_histogram.
|
num_histogram_buckets
|
An optional number of buckets in a standard
NumericStatistics.histogram with equal-width buckets.
|
num_quantiles_histogram_buckets
|
An optional number of buckets in a
quantiles NumericStatistics.histogram.
|
epsilon
|
An optional error tolerance for the computation of quantiles,
typically a small fraction close to zero (e.g. 0.01). Higher values of
epsilon increase the quantile approximation, and hence result in more
unequal buckets, but could improve performance, and resource
consumption.
|
infer_type_from_schema
|
A boolean to indicate whether the feature types
should be inferred from the schema. If set to True, an input schema must
be provided. This flag is used only when invoking TFDV through
tfdv.generate_statistics_from_csv.
|
desired_batch_size
|
An optional maximum number of examples to include in
each batch that is passed to the statistics generators. When invoking
TFDV using its end-to-end APIs (e.g.
generate_statistics_from_tfrecord), this option also controls the
decoder batch size -- if provided, the decoded RecordBatches that are to
be fed to TFDV will have the fixed batch size. When invoking TFDV using
tfdv.GenerateStatistics, this option only controls the maximum size of
RecordBatches constructed within StatsGenerators (a generator may
combine RecordBatches).
|
enable_semantic_domain_stats
|
If True statistics for semantic domains are
generated (e.g: image, text domains).
|
semantic_domain_stats_sample_rate
|
An optional sampling rate for semantic
domain statistics. If specified, semantic domain statistics is computed
over a sample.
|
per_feature_weight_override
|
If specified, the "example weight" paired
with a feature will be first looked up in this map and if not found,
fall back to weight_feature.
|
vocab_paths
|
An optional dictionary mapping vocab names to paths. Used in
the schema when specifying a NaturalLanguageDomain. The paths can either
be to GZIP-compressed TF record files that have a tfrecord.gz suffix or
to text files.
|
add_default_generators
|
Whether to invoke the default set of stats
generators in the run. Generators invoked consists of 1) the default
generators (controlled by this option); 2) user-provided generators (
controlled by the generators option); 3) semantic generators
(controlled by enable_semantic_domain_stats) and 4) schema-based
generators that are enabled based on information provided in the schema.
|
feature_allowlist
|
An optional list of names of the features to calculate
statistics for, or a list of paths.
|
experimental_use_sketch_based_topk_uniques
|
Deprecated, prefer
use_sketch_based_topk_uniques.
|
use_sketch_based_topk_uniques
|
if True, use the sketch based top-k and
uniques stats generator.
|
experimental_slice_functions
|
An optional list of functions that generate
slice keys for each example. Each slice function should take
pyarrow.RecordBatch as input and return an Iterable[Tuple[Text,
pyarrow.RecordBatch]]. Each tuple contains the slice key and the
corresponding sliced RecordBatch. Only one of
experimental_slice_functions or experimental_slice_sqls must be
specified.
|
experimental_slice_sqls
|
List of slicing SQL queries. The query must have
the following pattern: "SELECT STRUCT({feature_name} [AS {slice_key}])
[FROM example.feature_name [, example.feature_name, ... ] [WHERE ... ]]"
The “example.feature_name” inside the FROM statement is used to flatten
the repeated fields. For non-repeated fields, you can directly write the
query as follows: “SELECT STRUCT(non_repeated_feature_a,
non_repeated_feature_b)” In the query, the “example” is a key word that
binds to each input "row". The semantics of this variable will depend on
the decoding of the input data to the Arrow representation (e.g., for
tf.Example, each key is decoded to a separate column). Thus, structured
data can be readily accessed by iterating/unnesting the fields of the
"example" variable. Example 1: Slice on each value of a feature "SELECT
STRUCT(gender) FROM example.gender" Example 2: Slice on each value of
one feature and a specified value of another. "SELECT STRUCT(gender,
country) FROM example.gender, example.country WHERE country = 'USA'"
Only one of experimental_slice_functions or experimental_slice_sqls must
be specified.
|
experimental_result_partitions
|
The number of feature partitions to
combine output DatasetFeatureStatisticsLists into. If set to 1 (default)
output is globally combined. If set to value greater than one, up to
that many shards are returned, each containing a subset of features.
|
experimental_num_feature_partitions
|
If > 1, partitions computations by
supported generators to act on this many bundles of features. For best
results this should be set to at least several times less than the
number of features in a dataset, and never more than the available beam
parallelism.
|
slicing_config
|
an optional SlicingConfig. SlicingConfig includes
slicing_specs specified with feature keys, feature values or slicing SQL
queries.
|
experimental_filter_read_paths
|
If provided, tries to push down either
paths passed via feature_allowlist or via the schema (in that priority)
to the underlying read operation. Support depends on the file reader.
|
per_feature_stats_config
|
Supports granular control of what statistics are
enabled per feature. Experimental.
|
Construct an instance of stats options from a JSON representation.
Custom generators and slice_functions cannot being converted. As a result,
a ValueError will be raised when these options are specified and TFDV is
running in a setting where the stats options have been json-serialized,
first. This will happen in the case where TFDV is run as a TFX component.
The schema proto and slicing_config will be json_encoded.
View source on GitHub