Mixture of Experts

Gated mixture of specialist experts.

class modeva.models.MoMoERegressor(name: str = None, n_clusters: int = 10, cluster_method: str = 'kmeans', centroids: ndarray = None, calibration: bool = True, cv=3, feature_names: list = None, cluster_features: list = None, gating_features: list = None, ltc_baseline_n_estimators: int = 300, ltc_baseline_max_depth: int = 2, ltc_n_estimators: int = 100, ltc_max_depth: int = 5, expert: str = 'xgboost', *args, **kwargs)

A Mixture of Experts (MoE) regressor that combines multiple expert models for regression tasks.

This regressor creates a weighted ensemble of expert models, where each expert specializes in different regions of the input space. The final prediction is computed as a weighted sum of expert predictions, where weights are determined by a gating network based on the input features.

Parameters:

namestr, default=None

Identifier for the model instance.

n_clustersint, default=10

Number of expert models (clusters) to create.

cluster_method{‘ltc’, ‘kmeans’}, default=’kmeans’

Which algorithm to use.

• ‘kmeans’: Use KMeans clustering algorithm directly.

• ‘ltc’: This method (Learning Trajectory Cluster; LTC) first fits a baseline xgboost model and calculates its absolute residuals; then fits a gradient boosting models between predictors and the absolute residuals. It extracts prediction trajectories during training, applies optional weighting schemes, performs PCA dimensionality reduction, and clusters samples based on their learning patterns.

centroidsnp.ndarray, default=None

Pre-defined cluster centers of shape (n_clusters, n_features). If provided, skips the clustering step and uses these centers directly.

calibrationbool, default=True

Whether to calibrate the gating network’s probability estimates using cross-validation.

cvint, cross-validation generator or iterable, default=3

Cross-validation strategy for probability calibration. Can be: - int: number of folds for K-Fold cross-validation - cross-validation generator: custom splitting strategy - iterable: yields (train, test) splits as indices

feature_nameslist or None, default=None

The list of feature names. If None, will use “X0”, “X1”, “X2”, etc.

cluster_featureslist or None, default=None

The list of feature names used for clustering. If None, will use all features.

gating_featureslist or None, default=None

The list of feature names used for gating model fitting. If None, will use all features.

ltc_baseline_n_estimatorsint, default=300

The number of estimators used for fitting the baseline Xgboost, used when cluster_method=”ltc” only.

ltc_baseline_max_depthint, default=2

The max depth used for fitting the baseline Xgboost, used when cluster_method=”ltc” only.

ltc_n_estimatorsint, default=300

The number of estimators used for fitting the gradient boosting models in trajectory analysis, used when cluster_method=”ltc” only.

ltc_max_depthint, default=2

The max depth used for fitting the gradient boosting models in trajectory analysis, used when cluster_method=”ltc” only.

expert{“xgboost”, “lightgbm”, “catboost”}, default=”xgboost”

The backend estimator used for each cluster.

*args

Variable length argument list passed to the underlying expert model.

**kwargs

Arbitrary keyword arguments passed to the underlying expert model.

calibrate_interval(X, y, alpha=0.1, max_depth: int = 5)

Fit a conformal prediction model to the given data.

This method computes the model’s prediction interval calibrated to the given data.

If the model is a regressor, splits the data with 50% for fitting lower (alpha / 5) and upper (1 - alpha / 2) gradient boosting trees-based quantile regression to the model’s residual; and 50% for calibration.

If the model is a binary classifiers, it computes the calibration quantile based on predicted probabilities for the positive class.

Parameters:

XXnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

yarray-like of shape (n_samples, )

Target values.

alphafloat, default=0.1

Expected miscoverage for the conformal prediction.

max_depthint, default=5

Maximum depth of the gradient boosting trees for regression tasks. Only used when task_type is REGRESSION.

Raises:

ValueError: If the model is neither a regressor nor a classifier.

fit(X, y, sample_weight=None, **kwargs)

fit_cluster(cluster_id, cluster_labels, X, y, sample_weight, **kwargs)

Fit a model for a single cluster. This function is designed for parallelization.

get_metadata_routing()

Get metadata routing of this object.

Please check User Guide on how the routing mechanism works.

Returns:

routingMetadataRequest

A MetadataRequest encapsulating routing information.

get_params(deep=True)

Get parameters for this estimator.

Parameters:

deepbool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

paramsmapping of string to any

Parameter names mapped to their values.

get_pr_weight(X)

Return the weight of each expert.

get_raw_output(X)

Return the raw prediction outputs.

interpret(dataset)

load(file_name: str)

Load the model into memory from file system.

Parameters:

file_name: str

The path and name of the file.

Returns:

estimator object

predict(X)

Model predictions, calling the child class’s ‘_predict’ method.

Parameters:

Xnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

Returns:

np.ndarray: The (calibrated) final prediction

predict_interval(X)

Predict the prediction interval for the given data based on the conformal prediction model.

It splits the data with 50% for fitting lower (alpha / 5) and upper (1 - alpha / 2) gradient boosting trees-based quantile regression to the model’s residual; and 50% for calibration.

Parameters:

Xnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

Returns:

np.ndarray: The lower and upper bounds of the prediction intervals for each sample

in the format [n_samples, 2] for regressors or a flattened array for classifiers.

Raises:

ValueError: If fit_conformal has not been called to fit the conformal prediction model

before calling this method.

reset_calibrate_interval()

reset_calibrate_proba()

save(file_name: str)

Save the model into file system.

Parameters:

file_name: str

The path and name of the file.

score(X, y, sample_weight=None)

Return the coefficient of determination of the prediction.

The coefficient of determination \(R^2\) is defined as \((1 - \frac{u}{v})\), where \(u\) is the residual sum of squares ((y_true - y_pred)** 2).sum() and \(v\) is the total sum of squares ((y_true - y_true.mean()) ** 2).sum(). The best possible score is 1.0 and it can be negative (because the model can be arbitrarily worse). A constant model that always predicts the expected value of y, disregarding the input features, would get a \(R^2\) score of 0.0.

Parameters:

Xarray-like of shape (n_samples, n_features)

Test samples. For some estimators this may be a precomputed kernel matrix or a list of generic objects instead with shape (n_samples, n_samples_fitted), where n_samples_fitted is the number of samples used in the fitting for the estimator.

yarray-like of shape (n_samples,) or (n_samples, n_outputs)

True values for X.

sample_weightarray-like of shape (n_samples,), default=None

Sample weights.

Returns:

scorefloat

\(R^2\) of self.predict(X) w.r.t. y.

Notes

The \(R^2\) score used when calling score on a regressor uses multioutput='uniform_average' from version 0.23 to keep consistent with default value of r2_score(). This influences the score method of all the multioutput regressors (except for MultiOutputRegressor).

set_fit_request(*, sample_weight: bool | None | str = '$UNCHANGED$') → MoMoERegressor

Request metadata passed to the fit method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to fit if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to fit.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

sample_weightstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for sample_weight parameter in fit.

Returns:

selfobject

The updated object.

set_params(**params)

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as Pipeline). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters:

**paramsdict

Estimator parameters.

Returns:

selfestimator instance

Estimator instance.

set_score_request(*, sample_weight: bool | None | str = '$UNCHANGED$') → MoMoERegressor

Request metadata passed to the score method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to score if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to score.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

sample_weightstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for sample_weight parameter in score.

Returns:

selfobject

The updated object.

property name

property version

class modeva.models.MoMoEClassifier(name: str = None, n_clusters: int = 10, cluster_method: str = 'kmeans', centroids: ndarray = None, calibration: bool = True, cv=3, feature_names: list = None, cluster_features: list = None, gating_features: list = None, ltc_baseline_n_estimators: int = 300, ltc_baseline_max_depth: int = 2, ltc_n_estimators: int = 100, ltc_max_depth: int = 5, expert: str = 'xgboost', *args, **kwargs)

A Mixture of Experts (MoE) classifier that combines multiple expert models for classification tasks.

This classifier creates a weighted ensemble of expert models, where each expert specializes in different regions of the input space. The final prediction is computed by combining expert predictions weighted by the gating network’s outputs.

Parameters:

namestr, default=None

Identifier for the model instance.

n_clustersint, default=10

Number of expert models (clusters) to create.

cluster_method{‘ltc’, ‘kmeans’}, default=’kmeans’

Which algorithm to use.

• ‘kmeans’: Use KMeans clustering algorithm directly.

• ‘ltc’: This method (Learning Trajectory Cluster; LTC) first fits a baseline xgboost model and calculates its absolute residuals; then fits a gradient boosting models between predictors and the absolute residuals. It extracts prediction trajectories during training, applies optional weighting schemes, performs PCA dimensionality reduction, and clusters samples based on their learning patterns.

centroidsnp.ndarray, default=None

Pre-defined cluster centers of shape (n_clusters, n_features). If provided, skips the clustering step and uses these centers directly.

calibrationbool, default=True

Whether to calibrate the gating network’s probability estimates using cross-validation.

cvint, cross-validation generator or iterable, default=3

Cross-validation strategy for probability calibration. Can be: - int: number of folds for K-Fold cross-validation - cross-validation generator: custom splitting strategy - iterable: yields (train, test) splits as indices

feature_nameslist or None, default=None

The list of feature names. If None, will use “X0”, “X1”, “X2”, etc.

cluster_featureslist or None, default=None

The list of feature names used for clustering. If None, will use all features.

gating_featureslist or None, default=None

The list of feature names used for gating model fitting. If None, will use all features.

ltc_baseline_n_estimatorsint, default=300

The number of estimators used for fitting the baseline Xgboost, used when cluster_method=”ltc” only.

ltc_baseline_max_depthint, default=2

The max depth used for fitting the baseline Xgboost, used when cluster_method=”ltc” only.

ltc_n_estimatorsint, default=300

The number of estimators used for fitting the gradient boosting models in trajectory analysis, used when cluster_method=”ltc” only.

ltc_max_depthint, default=2

The max depth used for fitting the gradient boosting models in trajectory analysis, used when cluster_method=”ltc” only.

expert{“xgboost”, “lightgbm”, “catboost”}, default=”xgboost”

The backend estimator used for each cluster.

*args

Variable length argument list passed to the underlying expert model.

**kwargs

Arbitrary keyword arguments passed to the underlying expert model.

calibrate_interval(X, y, alpha=0.1)

Fit a conformal prediction model to the given data.

This method computes the model’s prediction interval calibrated to the given data.

It computes the calibration quantile based on predicted probabilities for the positive class.

Parameters:

XXnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

yarray-like of shape (n_samples, )

Target values.

alphafloat, default=0.1

Expected miscoverage for the conformal prediction.

Raises:

ValueError: If the model is neither a regressor nor a classifier.

calibrate_proba(X, y, sample_weight=None, method='sigmoid')

Fit the calibration method on the model’s predictions.

Parameters:

Xnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

ynp.ndarray of shape (n_samples, )

Ground truth labels.

sample_weightarray-like, shape (n_samples,), default=None

Sample weights.

method{‘sigmoid’, ‘isotonic’}, default=’sigmoid’

The calibration method.

• ‘sigmoid’: Platt’s method, i.e., fit a logistic regression on predicted probabilities and y

• ‘isotonic’: Fit an isotonic regression on predicted probabilities and y.

Returns:

self: Calibrated estimator

decision_function(X, calibration: bool = True)

Computes the decision function for the given input data.

Parameters:

Xnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

calibrationbool, default=True

If True, will use calibrated probability if calibration is done. Otherwise, will use raw probability.

Returns:

logit_predictionarray, shape (n_samples,) or (n_samples, n_classes)

Array of (calibrated) logit predictions.

fit(X, y, sample_weight=None, **kwargs)

fit_cluster(cluster_id, cluster_labels, X, y, sample_weight, **kwargs)

Fit a model for a single cluster. This function is designed for parallelization.

get_metadata_routing()

Get metadata routing of this object.

Please check User Guide on how the routing mechanism works.

Returns:

routingMetadataRequest

A MetadataRequest encapsulating routing information.

get_params(deep=True)

Get parameters for this estimator.

Parameters:

deepbool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

paramsmapping of string to any

Parameter names mapped to their values.

get_pr_weight(X)

Return the weight of each expert.

get_raw_output(X)

Return the raw prediction outputs.

interpret(dataset)

load(file_name: str)

Load the model into memory from file system.

Parameters:

file_name: str

The path and name of the file.

Returns:

estimator object

predict(X, calibration: bool = True)

Model predictions, calling the child class’s ‘_predict’ method.

Parameters:

Xnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

calibrationbool, default=True

If True, will use calibrated probability if calibration is done. Otherwise, will use raw probability.

Returns:

np.ndarray: The (calibrated) final prediction

predict_interval(X)

Predict the prediction set for the given data based on the conformal prediction model.

This method computes the model prediction interval (regression) or prediction sets (classification) using conformal prediction.

Parameters:

Xnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

Returns:

np.ndarray: The lower and upper bounds of the prediction intervals for each sample

in the format [n_samples, 2] for regressors or a flattened array for classifiers.

Raises:

ValueError: If fit_conformal has not been called to fit the conformal prediction model

before calling this method.

predict_proba(X, calibration: bool = True)

Predict (calibrated) probabilities for X.

Parameters:

Xnp.ndarray of shape (n_samples, n_features)

Feature matrix for prediction.

calibrationbool, default=True

If True, will return calibrated probability if calibration is done. Otherwise, will return raw probability.

Returns:

np.ndarray: The (calibrated) predicted probabilities

reset_calibrate_interval()

reset_calibrate_proba()

save(file_name: str)

Save the model into file system.

Parameters:

file_name: str

The path and name of the file.

score(X, y, sample_weight=None)

Return the mean accuracy on the given test data and labels.

In multi-label classification, this is the subset accuracy which is a harsh metric since you require for each sample that each label set be correctly predicted.

Parameters:

Xarray-like of shape (n_samples, n_features)

Test samples.

yarray-like of shape (n_samples,) or (n_samples, n_outputs)

True labels for X.

sample_weightarray-like of shape (n_samples,), default=None

Sample weights.

Returns:

scorefloat

Mean accuracy of self.predict(X) w.r.t. y.

set_decision_function_request(*, calibration: bool | None | str = '$UNCHANGED$') → MoMoEClassifier

Request metadata passed to the decision_function method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to decision_function if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to decision_function.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

calibrationstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for calibration parameter in decision_function.

Returns:

selfobject

The updated object.

set_fit_request(*, sample_weight: bool | None | str = '$UNCHANGED$') → MoMoEClassifier

Request metadata passed to the fit method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to fit if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to fit.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

sample_weightstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for sample_weight parameter in fit.

Returns:

selfobject

The updated object.

set_params(**params)

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as Pipeline). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters:

**paramsdict

Estimator parameters.

Returns:

selfestimator instance

Estimator instance.

set_predict_proba_request(*, calibration: bool | None | str = '$UNCHANGED$') → MoMoEClassifier

Request metadata passed to the predict_proba method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to predict_proba if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to predict_proba.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

calibrationstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for calibration parameter in predict_proba.

Returns:

selfobject

The updated object.

set_predict_request(*, calibration: bool | None | str = '$UNCHANGED$') → MoMoEClassifier

Request metadata passed to the predict method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to predict if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to predict.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

calibrationstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for calibration parameter in predict.

Returns:

selfobject

The updated object.

set_score_request(*, sample_weight: bool | None | str = '$UNCHANGED$') → MoMoEClassifier

Request metadata passed to the score method.

Note that this method is only relevant if enable_metadata_routing=True (see sklearn.set_config()). Please see User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to score if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to score.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Note

This method is only relevant if this estimator is used as a sub-estimator of a meta-estimator, e.g. used inside a Pipeline. Otherwise it has no effect.

Parameters:

sample_weightstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for sample_weight parameter in score.

Returns:

selfobject

The updated object.

property name

property version