Protocols

fyt.core.protocols

Protocols defining the interfaces for pipeline components.

These protocols decouple the TrainingPipeline orchestrator from concrete implementations, enabling substitution and easier testing.

DataSplitter

Bases: Protocol

Protocol for data loading and splitting.

split_data(test_size, random_state)

Split data into train/test sets.

Parameters:

Name	Type	Description	Default
test_size	float	Proportion of data to use for testing.	required
random_state	int	Random seed for reproducibility.	required

Returns:

Type	Description
DataSplit	DataSplit containing training and testing data.

FeatureFilter

Bases: Protocol

Protocol for feature selection.

fit(X, y=None)

Fit the feature selector on training data.

Parameters:

Name	Type	Description	Default
X	Any	Input features.	required
y	Any	Target variable (required for supervised methods).	None

Returns:

Type	Description
FeatureFilter	Fitted feature filter instance.

get_feature_names_out(input_features=None)

Get feature names after selection.

Parameters:

Name	Type	Description	Default
input_features	Any	Input feature names.	None

Returns:

Type	Description
ndarray	Array of selected feature names.

transform(X)

Transform data by selecting features.

Parameters:

Name	Type	Description	Default
X	Any	Input features.	required

Returns:

Type	Description
Any	Features after selection.

Predictor

Bases: Protocol

Protocol for model training and prediction.

best_params property writable

The best hyperparameters from optimization, if any.

model property

The trained model instance.

model_params property

The base model parameters.

model_type property

The model type identifier.

predict(X)

Make predictions.

Parameters:

Name	Type	Description	Default
X	DataFrame	Feature data for prediction.	required

Returns:

Type	Description
Series	Predicted values.

predict_proba(X)

Make probability predictions.

Parameters:

Name	Type	Description	Default
X	DataFrame	Feature data for prediction.	required

Returns:

Type	Description
DataFrame	Predicted probabilities.

train(data_split)

Train the model on training data.

Parameters:

Name	Type	Description	Default
data_split	DataSplit	Data split containing training data.	required

Preprocessor

Bases: Protocol

Protocol for feature preprocessing.

fit(X, y=None)

Fit the preprocessor on training data.

Parameters:

Name	Type	Description	Default
X	Any	Input features.	required
y	Any	Target variable (optional).	None

Returns:

Type	Description
Preprocessor	Fitted preprocessor instance.

handle_zero_imputation(X)

Handle zero-to-NaN conversion for configured features.

Parameters:

Name	Type	Description	Default
X	Any	Input features.	required

Returns:

Type	Description
Any	Features with zeros converted to NaN where configured.

transform(X)

Transform features using fitted preprocessor.

Parameters:

Name	Type	Description	Default
X	Any	Input features to transform.	required

Returns:

Type	Description
Any	Transformed features.

SeedAware

Bases: Protocol

Protocol for components that accept a random state seed.

TargetTransformer

Bases: Protocol

Protocol for target variable encoding.

class_mapping property

The class-to-label mapping, or None if not available.

fit_transform(y)

Fit the encoder and transform the target variable.

Parameters:

Name	Type	Description	Default
y	Series	Target variable to fit and transform.	required

Returns:

Type	Description
Series	Transformed target variable.

transform(y)

Transform the target variable using a fitted encoder.

Parameters:

Name	Type	Description	Default
y	Series	Target variable to transform.	required

Returns:

Type	Description
Series	Transformed target variable.