tanat.cast package

Submodules

tanat.cast.base module

Cast primitives, structural casts, and the recipe skeleton.

class tanat.cast.base.BaseCastRecipe

Bases: object

Common methods shared by sequence- and trajectory-level recipes.

Subclasses must:

• be decorated with @dataclass(frozen=True),

• declare structural: StructuralCasts and features: <FeatureCasts> with default factories,

• set _FEATURES_CLS to the dataclass used for features.

classmethod coerce(value)

Coerce value into a recipe instance.

Accepted:

• None → default recipe.

• a recipe instance → returned unchanged.

• a dict in the canonical nested form {"structural": ..., "features": ...} (each side may itself be a dict, which is then unpacked into the matching dataclass).

copy()

Return a deep copy of the recipe.

features: object

property id: list[DataType]

Ordered dtype steps of the id recipe.

id_caster() → Callable[[Expr], Expr] | None

Return the compiled caster for the id column, or None.

property id_dtype: DataType | None

Final dtype of the id recipe, None if no cast is registered.

is_empty() → bool

Return True when no structural and no feature cast is registered.

replace(**kwargs)

Functional update.

Recognised keys:

• structural= / features=: canonical nested form.

• id= : shortcut to reset the structural id recipe.

• time_index=: shortcut to reset the structural time_index recipe.

property static: dict[str, list[DataType]]

Per-column dtype steps of the static feature recipes.

static_caster(col: str) → Callable[[Expr], Expr] | None

Return the compiled caster for static feature col, or None.

structural: StructuralCasts

property time_index: list[DataType]

Ordered dtype steps of the time_index recipe.

time_index_caster() → Callable[[Expr], Expr] | None

Return the compiled caster for the time_index column, or None.

property time_index_dtype: DataType | None

Final dtype of the time_index recipe, None if no cast is registered.

class tanat.cast.base.ColumnMapCast(recipes: dict[str, list[~polars.datatypes.classes.DataType]]=<factory>)

Bases: object

Ordered cast recipes for a named set of columns.

__init__(recipes: dict[str, list[~polars.datatypes.classes.DataType]]=<factory>) → None

append(schema: dict[str, DataType]) → ColumnMapCast

Return a new map with each col → dtype of schema appended.

apply(lf: LazyFrame) → LazyFrame

Apply all registered column casts to lf (no-op when empty).

caster(col: str) → Callable[[Expr], Expr] | None

Return the compiled caster for col, or None when unset.

copy() → ColumnMapCast

Return a deep copy of the map.

exprs() → list[Expr]

Return one with_columns expression per registered column.

is_empty() → bool

Return True when no column has a registered cast recipe.

probe_lf(lf: LazyFrame, n_rows: int = 10) → None

Eagerly evaluate an n_rows sample of lf to surface cast errors.

The candidate recipe is applied only for the probe; lf itself remains lazy and unchanged for the caller.

recipes: dict[str, list[DataType]]

class tanat.cast.base.ScalarCast(recipe: list[DataType] = <factory>)

Bases: object

Ordered cast recipe for a single, well-known column.

__init__(recipe: list[DataType] = <factory>) → None

append(dtype: DataType) → ScalarCast

Return a new recipe with dtype appended as a final step.

caster() → Callable[[Expr], Expr] | None

Return the compiled caster, or None when the recipe is empty.

copy() → ScalarCast

Return a deep copy of the recipe.

final_dtype() → DataType | None

Return the dtype the recipe ultimately casts to, None if empty.

is_empty() → bool

Return True when no cast step is registered.

recipe: list[DataType]

class tanat.cast.base.StructuralCasts(id: ScalarCast = <factory>, time_index: ScalarCast = <factory>)

Bases: object

Shared ordered casts consumed by stores: id and time index.

__init__(id: ScalarCast = <factory>, time_index: ScalarCast = <factory>) → None

append_id(dtype: DataType) → StructuralCasts

Append dtype to the id recipe; reject closed-domain dtypes.

pl.Enum / pl.Categorical would conflict with any id outside the current view (the store consumes the cast on the full physical data). Materialise the view via save() first to narrow the domain.

append_time_index(dtype: DataType) → StructuralCasts

Append dtype to the time_index recipe.

apply(lf: LazyFrame, *, id_col: str | None = None, time_cols: list[str] | None = None) → LazyFrame

Apply id and (optionally) time-index structural casts to lf.

Parameters:

• lf – Input LazyFrame.

• id_col – Name of the id column to cast. None → skip.

• time_cols – Names of time columns to cast. None / [] → skip.

copy() → StructuralCasts

Return a deep copy of the structural casts.

id: ScalarCast

id_caster() → Callable[[Expr], Expr] | None

Return the compiled caster for the id column, or None.

is_empty() → bool

Return True when both id and time_index recipes are empty.

probe(view) → None

Validate id and time_index recipes against full structural store data.

time_index: ScalarCast

time_index_caster() → Callable[[Expr], Expr] | None

Return the compiled caster for the time_index column, or None.

tanat.cast.base.apply_cast_exprs(lf: LazyFrame, exprs: list[Expr]) → LazyFrame

Apply pre-built cast expressions to lf.

Returns lf unchanged when exprs is empty.

tanat.cast.base.build_caster(recipe: list[DataType]) → Callable[[Expr], Expr]

Build an expression caster from an ordered dtype recipe.

tanat.cast.base.probe_cast_recipe(lf: LazyFrame, recipes: dict[str, list[DataType]], n_rows: int = 10) → None

Validate multi-step cast recipes on a small sample of lf.

Each column is cast through its recipe in order (T0 → T1 → … → Tn). Absent columns are silently skipped.

Raises:

TypeError – If any step fails, with column name and full chain in the message.

tanat.cast.sequence module

Sequence-level cast recipes.

class tanat.cast.sequence.SequenceCastRecipe(structural: StructuralCasts = <factory>, features: SequenceFeatureCasts = <factory>)

Bases: BaseCastRecipe

Type-cast overrides for a sequence view.

__init__(structural: StructuralCasts = <factory>, features: SequenceFeatureCasts = <factory>) → None

append(*, id: DataType | None = None, time_index: DataType | None = None, entity: dict[str, DataType] | None = None, static: dict[str, DataType] | None = None) → SequenceCastRecipe

Return a new recipe extended with the given structural and feature casts.

property entity: dict[str, list[DataType]]

Per-column dtype steps of the entity feature recipes.

entity_caster(col: str) → Callable[[Expr], Expr] | None

Return the compiled caster for entity feature col, or None.

feature_exprs(is_static: bool = False) → list[Expr]

Return the with_columns expressions for entity or static features.

features: SequenceFeatureCasts

fork_casts(*, feature_col: str | None = None, static_col: str | None = None) → _ForkCasts

Return precompiled casts for a fork operation.

probe(view) → None

Validate structural and all feature casts against view.

structural: StructuralCasts

class tanat.cast.sequence.SequenceFeatureCasts(entity: ColumnMapCast = <factory>, static: ColumnMapCast = <factory>)

Bases: object

Feature casts consumed by sequence views.

__init__(entity: ColumnMapCast = <factory>, static: ColumnMapCast = <factory>) → None

apply(lf: LazyFrame, *, is_static: bool) → LazyFrame

Apply entity or static feature casts to lf (no-op when empty).

copy() → SequenceFeatureCasts

Return a deep copy of the feature casts.

entity: ColumnMapCast

is_empty() → bool

Return True when no entity and no static cast is registered.

probe(view, *, is_static: bool) → None

Probe this candidate recipe on top of the current sequence view.

probe_lf(lf: LazyFrame | None, *, is_static: bool) → None

Force-evaluate the selected bucket on a view LazyFrame.

static: ColumnMapCast

tanat.cast.trajectory module

Trajectory-level cast recipes.

class tanat.cast.trajectory.TrajectoryCastRecipe(structural: StructuralCasts = <factory>, features: TrajectoryFeatureCasts = <factory>)

Bases: BaseCastRecipe

Type-cast overrides for a trajectory view.

Entity-level feature casts live in the trajectory’s sub-pools (see SequenceCastRecipe); at the trajectory layer only static features are observable directly.

__init__(structural: StructuralCasts = <factory>, features: TrajectoryFeatureCasts = <factory>) → None

append(*, id: DataType | None = None, time_index: DataType | None = None, static: dict[str, DataType] | None = None) → TrajectoryCastRecipe

Return a new recipe extended with the given structural and static casts.

feature_exprs() → list[Expr]

Return the with_columns expressions for static features.

features: TrajectoryFeatureCasts

probe(view) → None

Validate structural and static feature casts against view.

structural: StructuralCasts

class tanat.cast.trajectory.TrajectoryFeatureCasts(static: ColumnMapCast = <factory>)

Bases: object

Feature casts consumed by trajectory views (static only).

__init__(static: ColumnMapCast = <factory>) → None

apply(lf: LazyFrame) → LazyFrame

Apply static feature casts to lf (no-op when empty).

copy() → TrajectoryFeatureCasts

Return a deep copy of the feature casts.

is_empty() → bool

Return True when no static cast is registered.

probe(view) → None

Probe this candidate recipe on top of the current trajectory view.

probe_lf(lf: LazyFrame | None) → None

Force-evaluate the static bucket on a view LazyFrame.

static: ColumnMapCast

Module contents

Cast recipes and reusable cast primitives.

class tanat.cast.ColumnMapCast(recipes: dict[str, list[~polars.datatypes.classes.DataType]]=<factory>)

Bases: object

Ordered cast recipes for a named set of columns.

__init__(recipes: dict[str, list[~polars.datatypes.classes.DataType]]=<factory>) → None

append(schema: dict[str, DataType]) → ColumnMapCast

Return a new map with each col → dtype of schema appended.

apply(lf: LazyFrame) → LazyFrame

Apply all registered column casts to lf (no-op when empty).

caster(col: str) → Callable[[Expr], Expr] | None

Return the compiled caster for col, or None when unset.

copy() → ColumnMapCast

Return a deep copy of the map.

exprs() → list[Expr]

Return one with_columns expression per registered column.

is_empty() → bool

Return True when no column has a registered cast recipe.

probe_lf(lf: LazyFrame, n_rows: int = 10) → None

Eagerly evaluate an n_rows sample of lf to surface cast errors.

The candidate recipe is applied only for the probe; lf itself remains lazy and unchanged for the caller.

recipes: dict[str, list[DataType]]

class tanat.cast.ScalarCast(recipe: list[DataType] = <factory>)

Bases: object

Ordered cast recipe for a single, well-known column.

__init__(recipe: list[DataType] = <factory>) → None

append(dtype: DataType) → ScalarCast

Return a new recipe with dtype appended as a final step.

caster() → Callable[[Expr], Expr] | None

Return the compiled caster, or None when the recipe is empty.

copy() → ScalarCast

Return a deep copy of the recipe.

final_dtype() → DataType | None

Return the dtype the recipe ultimately casts to, None if empty.

is_empty() → bool

Return True when no cast step is registered.

recipe: list[DataType]

class tanat.cast.SequenceCastRecipe(structural: StructuralCasts = <factory>, features: SequenceFeatureCasts = <factory>)

Bases: BaseCastRecipe

Type-cast overrides for a sequence view.

__init__(structural: StructuralCasts = <factory>, features: SequenceFeatureCasts = <factory>) → None

append(*, id: DataType | None = None, time_index: DataType | None = None, entity: dict[str, DataType] | None = None, static: dict[str, DataType] | None = None) → SequenceCastRecipe

Return a new recipe extended with the given structural and feature casts.

property entity: dict[str, list[DataType]]

Per-column dtype steps of the entity feature recipes.

entity_caster(col: str) → Callable[[Expr], Expr] | None

Return the compiled caster for entity feature col, or None.

feature_exprs(is_static: bool = False) → list[Expr]

Return the with_columns expressions for entity or static features.

features: SequenceFeatureCasts

fork_casts(*, feature_col: str | None = None, static_col: str | None = None) → _ForkCasts

Return precompiled casts for a fork operation.

probe(view) → None

Validate structural and all feature casts against view.

structural: StructuralCasts

class tanat.cast.SequenceFeatureCasts(entity: ColumnMapCast = <factory>, static: ColumnMapCast = <factory>)

Bases: object

Feature casts consumed by sequence views.

__init__(entity: ColumnMapCast = <factory>, static: ColumnMapCast = <factory>) → None

apply(lf: LazyFrame, *, is_static: bool) → LazyFrame

Apply entity or static feature casts to lf (no-op when empty).

copy() → SequenceFeatureCasts

Return a deep copy of the feature casts.

entity: ColumnMapCast

is_empty() → bool

Return True when no entity and no static cast is registered.

probe(view, *, is_static: bool) → None

Probe this candidate recipe on top of the current sequence view.

probe_lf(lf: LazyFrame | None, *, is_static: bool) → None

Force-evaluate the selected bucket on a view LazyFrame.

static: ColumnMapCast

class tanat.cast.StructuralCasts(id: ScalarCast = <factory>, time_index: ScalarCast = <factory>)

Bases: object

Shared ordered casts consumed by stores: id and time index.

__init__(id: ScalarCast = <factory>, time_index: ScalarCast = <factory>) → None

append_id(dtype: DataType) → StructuralCasts

Append dtype to the id recipe; reject closed-domain dtypes.

pl.Enum / pl.Categorical would conflict with any id outside the current view (the store consumes the cast on the full physical data). Materialise the view via save() first to narrow the domain.

append_time_index(dtype: DataType) → StructuralCasts

Append dtype to the time_index recipe.

apply(lf: LazyFrame, *, id_col: str | None = None, time_cols: list[str] | None = None) → LazyFrame

Apply id and (optionally) time-index structural casts to lf.

Parameters:

• lf – Input LazyFrame.

• id_col – Name of the id column to cast. None → skip.

• time_cols – Names of time columns to cast. None / [] → skip.

copy() → StructuralCasts

Return a deep copy of the structural casts.

id: ScalarCast

id_caster() → Callable[[Expr], Expr] | None

Return the compiled caster for the id column, or None.

is_empty() → bool

Return True when both id and time_index recipes are empty.

probe(view) → None

Validate id and time_index recipes against full structural store data.

time_index: ScalarCast

time_index_caster() → Callable[[Expr], Expr] | None

Return the compiled caster for the time_index column, or None.

class tanat.cast.TrajectoryCastRecipe(structural: StructuralCasts = <factory>, features: TrajectoryFeatureCasts = <factory>)

Bases: BaseCastRecipe

Type-cast overrides for a trajectory view.

Entity-level feature casts live in the trajectory’s sub-pools (see SequenceCastRecipe); at the trajectory layer only static features are observable directly.

__init__(structural: StructuralCasts = <factory>, features: TrajectoryFeatureCasts = <factory>) → None

append(*, id: DataType | None = None, time_index: DataType | None = None, static: dict[str, DataType] | None = None) → TrajectoryCastRecipe

Return a new recipe extended with the given structural and static casts.

feature_exprs() → list[Expr]

Return the with_columns expressions for static features.

features: TrajectoryFeatureCasts

probe(view) → None

Validate structural and static feature casts against view.

structural: StructuralCasts

class tanat.cast.TrajectoryFeatureCasts(static: ColumnMapCast = <factory>)

Bases: object

Feature casts consumed by trajectory views (static only).

__init__(static: ColumnMapCast = <factory>) → None

apply(lf: LazyFrame) → LazyFrame

Apply static feature casts to lf (no-op when empty).

copy() → TrajectoryFeatureCasts

Return a deep copy of the feature casts.

is_empty() → bool

Return True when no static cast is registered.

probe(view) → None

Probe this candidate recipe on top of the current trajectory view.

probe_lf(lf: LazyFrame | None) → None

Force-evaluate the static bucket on a view LazyFrame.

static: ColumnMapCast

tanat.cast.apply_cast_exprs(lf: LazyFrame, exprs: list[Expr]) → LazyFrame

Apply pre-built cast expressions to lf.

Returns lf unchanged when exprs is empty.

tanat.cast.probe_cast_recipe(lf: LazyFrame, recipes: dict[str, list[DataType]], n_rows: int = 10) → None

Validate multi-step cast recipes on a small sample of lf.

Each column is cast through its recipe in order (T0 → T1 → … → Tn). Absent columns are silently skipped.

Raises:

TypeError – If any step fails, with column name and full chain in the message.