ONN

class ordinal_xai.models.onn.ONN(hidden_layers: ~typing.List[int] = [64, 64], output_layer: ~torch.nn.modules.module.Module = <class 'dlordinal.output_layers.stick_breaking_layer.StickBreakingLayer'>, loss_function: ~torch.nn.modules.module.Module = <class 'dlordinal.losses.cdw.CDWCELoss'>, dropout: float = 0.2, max_epochs: int = 1000, batch_size: int = 32, lr: float = 0.0001, optimizer: ~torch.optim.optimizer.Optimizer = <class 'torch.optim.adam.Adam'>, patience: int = 10, min_delta: float = 0.0001, verbose: int = 2)[source]

Bases: BaseEstimator, BaseOrdinalModel

Ordinal Neural Network for ordinal regression.

This class implements an ordinal regression model using a neural network with a triangular loss function. The model uses a fully connected network with configurable architecture and training parameters.

Parameters:

  • hidden_layers (List[int], default=[64, 64]) – List of hidden layer sizes

  • dropout (float, default=0.2) – Dropout rate for regularization

  • max_epochs (int, default=10000) – Maximum number of training epochs

  • batch_size (int, default=32) – Batch size for training

  • lr (float, default=0.001) – Learning rate for optimization

  • patience (int, default=10) – Number of epochs to wait for improvement before early stopping

  • min_delta (float, default=0.0001) – Minimum change in validation loss to be considered as improvement

  • verbose (int, default=2) – Verbosity level for training progress

feature_names_

Names of features used during training

Type:

list

n_features_in_

Number of features seen during training

Type:

int

ranks_

Unique ordinal class labels

Type:

ndarray

_model

The fitted skorch neural network

Type:

NeuralNet

_encoder

Encoder for categorical features

Type:

OneHotEncoder

_scaler

Scaler for numerical features

Type:

StandardScaler

is_fitted_

Whether the model has been fitted

Type:

bool

Notes

  • The model handles both categorical and numerical features automatically

  • Categorical features are one-hot encoded

  • Numerical features are standardized

  • Uses early stopping to prevent overfitting

  • Automatically uses GPU if available

__init__(hidden_layers: ~typing.List[int] = [64, 64], output_layer: ~torch.nn.modules.module.Module = <class 'dlordinal.output_layers.stick_breaking_layer.StickBreakingLayer'>, loss_function: ~torch.nn.modules.module.Module = <class 'dlordinal.losses.cdw.CDWCELoss'>, dropout: float = 0.2, max_epochs: int = 1000, batch_size: int = 32, lr: float = 0.0001, optimizer: ~torch.optim.optimizer.Optimizer = <class 'torch.optim.adam.Adam'>, patience: int = 10, min_delta: float = 0.0001, verbose: int = 2)[source]

Initialize the Ordinal Neural Network.

Parameters:

  • hidden_layers (List[int], default=[64, 64]) – List of hidden layer sizes

  • dropout (float, default=0.2) – Dropout rate for regularization

  • max_epochs (int, default=1000) – Maximum number of training epochs

  • batch_size (int, default=32) – Batch size for training

  • lr (float, default=0.001) – Learning rate for optimization

  • optimizer (torch.optim.Optimizer, default=torch.optim.Adam) – Optimizer for optimization

  • patience (int, default=10) – Number of epochs to wait for improvement before early stopping

  • min_delta (float, default=0.0001) – Minimum change in validation loss to be considered as improvement

  • verbose (int, default=2) – Verbosity level for training progress

get_params(deep: bool = True) → Dict[str, any][source]

Get parameters for this estimator.

Parameters:

deep (bool, default=True) – If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

Parameter names mapped to their values

Return type:

dict

set_params(**params: any) → ONN[source]

Set the parameters of this estimator.

Parameters:

**params (dict) – Estimator parameters

Returns:

self – The estimator instance

Return type:

ONN

fit(X: DataFrame, y: Series) → ONN[source]

Fit the Ordinal Neural Network model.

This method fits the model to the training data, handling both categorical and numerical features appropriately. It uses early stopping to prevent overfitting and automatically selects the best device (CPU/GPU) for training.

Parameters:

  • X (pd.DataFrame of shape (n_samples, n_features)) – Training data

  • y (pd.Series of shape (n_samples,)) – Target values

Returns:

self – The fitted model

Return type:

ONN

Raises:

  • ValueError – If the input data contains invalid values

  • RuntimeError – If there are issues with the neural network training

predict(X: DataFrame) → ndarray[source]

Predict ordinal class labels.

Parameters:

X (pd.DataFrame of shape (n_samples, n_features)) – Samples to predict

Returns:

Predicted ordinal class labels

Return type:

ndarray of shape (n_samples,)

Raises:

NotFittedError – If the model has not been fitted

predict_proba(X: DataFrame) → ndarray[source]

Predict class probabilities.

Parameters:

X (pd.DataFrame of shape (n_samples, n_features)) – Samples to predict probabilities for

Returns:

Predicted class probabilities

Return type:

ndarray of shape (n_samples, n_classes)

Raises:

NotFittedError – If the model has not been fitted

transform(X: DataFrame, fit: bool = False, no_scaling: bool = False) → DataFrame[source]

Transform input data into the format expected by the model.

This method handles both categorical and numerical features: - Categorical features are one-hot encoded - Numerical features are standardized (unless no_scaling=True)

Parameters:

  • X (pd.DataFrame of shape (n_samples, n_features)) – Input data to transform

  • fit (bool, default=False) – Whether to fit new encoder/scaler or use existing ones

  • no_scaling (bool, default=False) – Whether to skip scaling of numerical features

Returns:

Transformed data

Return type:

pd.DataFrame

Raises:

ValueError – If the input data has different features than training data

_abc_impl = <_abc._abc_data object>
set_transform_request(*, fit: bool | None | str = '$UNCHANGED$', no_scaling: bool | None | str = '$UNCHANGED$') → ONN

Configure whether metadata should be requested to be passed to the transform method.

Note that this method is only relevant when this estimator is used as a sub-estimator within a meta-estimator and metadata routing is enabled with enable_metadata_routing=True (see sklearn.set_config()). Please check the User Guide on how the routing mechanism works.

The options for each parameter are:

  • True: metadata is requested, and passed to transform 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 transform.

  • 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.

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

Metadata routing for fit parameter in transform.

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

Metadata routing for no_scaling parameter in transform.

selfobject

The updated object.