FeaturePolicy#

class scikitplot.feature_selection.FeaturePolicy(roles=(), kinds=(), tags=(), assessments=(), overrides=(), binning=<factory>, drift=<factory>, on_missing_column='raise')[source]#

Complete, declarative description of how to profile a table.

Parameters:

rolesmapping of role to ColumnRule-like, default=(): Columns with a role other than "feature" (target, drop, identifier, weight, group, time). A column matched by two roles is an error.
kindsmapping of column to kind, default=(): Explicit semantic kinds that replace inference (for example a numeric zip_code declared "categorical_nominal", or "text").
tagsmapping of tag to ColumnRule-like, default=(): Free user labels ("one_hot", "engineered", "sensitive"); a column can carry many tags. Tags never change inference.
assessmentsmapping of name to AssessmentSpec-like, default=(): Merged over DEFAULT_ASSESSMENTS by name.
overridesmapping of column to mapping of assessment to bool, default=(): Manual answers. The inferred answer is kept next to the override.
binningBinningPolicy or mapping, default=BinningPolicy(): Bin recommendation settings.
driftDriftPolicy or mapping, default=DriftPolicy(): Drift reference settings.
on_missing_column{“raise”, “ignore”}, default=”raise”: What to do when roles, kinds, tags or overrides name a column that is not in the data.

Parameters:

roles (tuple[tuple[str, ColumnRule], ...])
kinds (tuple[tuple[str, str], ...])
tags (tuple[tuple[str, ColumnRule], ...])
assessments (tuple[tuple[str, AssessmentSpec], ...])
overrides (tuple[tuple[str, tuple[tuple[str, bool], ...]], ...])
binning (BinningPolicy)
drift (DriftPolicy)
on_missing_column (str)

See also

load_policy: Read a policy from TOML, YAML or JSON.
dump_policy: Write a policy, verified by reading it back.
profile_features: Fit a schema from a DataFrame and a policy.

Notes

User note. Keep one policy per project in version control. Its digest is recorded in every fitted schema, so you can tell which configuration produced a result.

Developer note. Stored containers are sorted tuples of pairs; the *_map properties return fresh dictionaries.

Examples

>>> from scikitplot.feature_selection import FeaturePolicy
>>> policy = FeaturePolicy(
...     roles={"target": ["churn"]},
...     assessments={
...         "skewed": {"provider": "skewed", "params": {"threshold": 0.75}}
...     },
... )
>>> policy.assessments_map["skewed"].params_dict["threshold"]
0.75
>>> FeaturePolicy.from_dict(policy.to_dict()) == policy
True

assessments: tuple[tuple[str, AssessmentSpec], ...] = ()#

property assessments_map: dict[str, AssessmentSpec]#

Assessment name to spec, defaults included.

Type:: dict

binning: BinningPolicy[source]#

check_columns(names)[source]#

Refuse kinds/overrides keys absent from names (unless ignored).

Parameters:

namessequence of str: Column names of the data.

Raises:

PolicyError: If a key is missing and on_missing_column="raise".

Parameters:

names (Sequence[str])

Return type:

None

property digest: str#

SHA-256 of the canonical JSON form; equal policies share it.

Type:: str

drift: DriftPolicy[source]#

dumps(format='toml')[source]#

Serialise the policy, verified by reading it back.

Parameters:

format{“toml”, “yaml”, “json”}, default=”toml”: Output format.

Returns:

str: The document.

Raises:

PolicyError: If the document would not load back equal to this policy.

Parameters:

format (str)

Return type:

str

classmethod from_dict(data)[source]#

Build a policy from its config form, checking every field first.

Parameters:

datamapping: A document as produced by to_dict or written by hand; every key is optional except that schema_version, when given, must be the supported generation.

Returns:

FeaturePolicy: The policy.

Raises:

PolicyError: For an unknown key, a wrong type or an out-of-range value.
IncompatibleSchemaError: For another schema_version.

Parameters:

data (Mapping[str, Any])

Return type:

FeaturePolicy

kinds: tuple[tuple[str, str], ...] = ()#

property kinds_map: dict[str, str]#

Column to declared kind.

Type:: dict

classmethod loads(text, format='toml')[source]#

Parse a policy document.

Parameters:

textstr: The document.
format{“toml”, “yaml”, “json”}, default=”toml”: Input format.

Returns:

FeaturePolicy: The policy.

Parameters:

text (str)
format (str)

Return type:

FeaturePolicy

on_missing_column: str = 'raise'#

overrides: tuple[tuple[str, tuple[tuple[str, bool], ...]], ...] = ()#

property overrides_map: dict[str, dict[str, bool]]#

Column to manual assessment answers.

Type:: dict

resolve_roles(names)[source]#

Return the role of every column in names.

Parameters:

namessequence of str: Column names of the data.

Returns:

dict: Column to role value; unlisted columns are "feature".

Raises:

PolicyError: If a column matches two roles, or a named column is missing and on_missing_column="raise".

Parameters:

names (Sequence[str])

Return type:

dict[str, str]

resolve_tags(names)[source]#

Return the tags of every column in names.

Parameters:

namessequence of str: Column names of the data.

Returns:

dict: Column to sorted tuple of tags (empty when untagged).

Parameters:

names (Sequence[str])

Return type:

dict[str, tuple[str, …]]

roles: tuple[tuple[str, ColumnRule], ...] = ()#

property roles_map: dict[str, ColumnRule]#

Role name to column rule.

Type:: dict

tags: tuple[tuple[str, ColumnRule], ...] = ()#

property tags_map: dict[str, ColumnRule]#

Tag to column rule.

Type:: dict

to_dict()[source]#

Return the canonical config form, defaults included.

Returns:

dict: A tree of the shared data model with schema_version.

Return type:

dict[str, Any]