sklearn_utilities.torch.skorch package

class sklearn_utilities.torch.skorch.AlgebraicErrors(*args, **kwargs)

Bases: Module

Algebraic errors. Loss = x^2 / sqrt(1 + x^2)

See also

https

//github.com/tuantle/regression-losses-pytorch

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.AllowNan(loss: Module, *, has_nan: Literal['left', 'right', 'both'] = 'both')

Bases: Module

Replaces NaNs in the target values with the predictions.

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.AsymmetricLoss(*, t: float, loss_pred_grater: Module = LNErrors(), loss_pred_less: Module = LNErrors())

Bases: Module

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.AsymmetricLosses(*, ts: Sequence[float] | int, loss: nn.Module = LNErrors())

Bases: Module

Asymmetric loss with multioutput support.

Parameters:

• y_pred (torch.Tensor) – The predicted values. [B, Ts * Ny] or [B, Ts, Ny] or [B, Ny, Ts]

• y_true (torch.Tensor) – The true values. [B, Ny] or [B] (if Ny == 1)

Returns:

0-dim tensor with the loss.

Return type:

torch.Tensor

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.LNErrors(n: int = 2)

Bases: Module

Returns L^n errors (not mean of L^n errors).

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.LogCoshErrors(*, softplus: bool = True, eps: float | None = None)

Bases: Module

Log cosh errors. Loss = log(cosh(errors + eps))

See also

https

//datascience.stackexchange.com/questions/96271/logcoshloss-on-pytorch

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.SkorchCNNReshaper(estimator: TEstimator, *, window_size: int | None)

Bases: EstimatorWrapperBase[TEstimator], Generic[TEstimator]

skorch wrapper that reshapes tabular data for CNNs using sliding windows.

estimator: TEstimator

fit(X: TX, y: TY, **fit_params: Any) → Self

predict(X: TX, **predict_params: Any) → TY

class sklearn_utilities.torch.skorch.SkorchCNNReshaperProba(estimator: TEstimator, *, window_size: int | None, m_type: Literal['mean', 'median', 'nanmean', 'nanmedian'] = 'mean', var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd'] = 'std')

Bases: SkorchReshaperProbaMixin[TEstimator], SkorchCNNReshaper[TEstimator], Generic[TEstimator]

estimator: TEstimator

m_type: Literal['mean', 'median', 'nanmean', 'nanmedian']

M-statistics type to return from predict by default

set_predict_request(*, return_std: bool | None | str = '$UNCHANGED$', type_: bool | None | str = '$UNCHANGED$') → SkorchCNNReshaperProba

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.

New 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:

• return_std (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for return_std parameter in predict.

• type (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for type_ parameter in predict.

Returns:

self – The updated object.

Return type:

object

var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd']

Variance type to return from predict by default

class sklearn_utilities.torch.skorch.SkorchReshaper(estimator: TEstimator)

Bases: EstimatorWrapperBase[TEstimator], Generic[TEstimator]

skorch wrapper that reshapes tabular data for NNs.

estimator: TEstimator

fit(X: TX, y: TY, **fit_params: Any) → Self

predict(X: TX, **predict_params: Any) → TY

class sklearn_utilities.torch.skorch.SkorchReshaperProba(estimator: TEstimator, m_type: Literal['mean', 'median', 'nanmean', 'nanmedian'] = 'mean', var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd'] = 'std')

Bases: SkorchReshaperProbaMixin[TEstimator], SkorchReshaper[TEstimator], Generic[TEstimator]

estimator: TEstimator

m_type: Literal['mean', 'median', 'nanmean', 'nanmedian']

M-statistics type to return from predict by default

set_predict_request(*, return_std: bool | None | str = '$UNCHANGED$', type_: bool | None | str = '$UNCHANGED$') → SkorchReshaperProba

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.

New 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:

• return_std (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for return_std parameter in predict.

• type (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for type_ parameter in predict.

Returns:

self – The updated object.

Return type:

object

var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd']

Variance type to return from predict by default

class sklearn_utilities.torch.skorch.XSigmoidErrors(*args, **kwargs)

Bases: Module

XSigmoid errors. Loss = x * (2 * sigmoid(x) - 1)

See also

https

//github.com/tuantle/regression-losses-pytorch

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.XTanhErrors(*args, **kwargs)

Bases: Module

XTanh errors. Loss = x * tanh(x)

See also

https

//github.com/tuantle/regression-losses-pytorch

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

Submodules

sklearn_utilities.torch.skorch.proba module

class sklearn_utilities.torch.skorch.proba.AlgebraicErrors(*args, **kwargs)

Bases: Module

Algebraic errors. Loss = x^2 / sqrt(1 + x^2)

See also

https

//github.com/tuantle/regression-losses-pytorch

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.proba.AllowNan(loss: Module, *, has_nan: Literal['left', 'right', 'both'] = 'both')

Bases: Module

Replaces NaNs in the target values with the predictions.

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.proba.AsymmetricLoss(*, t: float, loss_pred_grater: Module = LNErrors(), loss_pred_less: Module = LNErrors())

Bases: Module

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.proba.AsymmetricLosses(*, ts: Sequence[float] | int, loss: nn.Module = LNErrors())

Bases: Module

Asymmetric loss with multioutput support.

Parameters:

• y_pred (torch.Tensor) – The predicted values. [B, Ts * Ny] or [B, Ts, Ny] or [B, Ny, Ts]

• y_true (torch.Tensor) – The true values. [B, Ny] or [B] (if Ny == 1)

Returns:

0-dim tensor with the loss.

Return type:

torch.Tensor

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.proba.LNErrors(n: int = 2)

Bases: Module

Returns L^n errors (not mean of L^n errors).

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.proba.LogCoshErrors(*, softplus: bool = True, eps: float | None = None)

Bases: Module

Log cosh errors. Loss = log(cosh(errors + eps))

See also

https

//datascience.stackexchange.com/questions/96271/logcoshloss-on-pytorch

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.proba.SkorchCNNReshaperProba(estimator: TEstimator, *, window_size: int | None, m_type: Literal['mean', 'median', 'nanmean', 'nanmedian'] = 'mean', var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd'] = 'std')

Bases: SkorchReshaperProbaMixin[TEstimator], SkorchCNNReshaper[TEstimator], Generic[TEstimator]

estimator: TEstimator

m_type: Literal['mean', 'median', 'nanmean', 'nanmedian']

M-statistics type to return from predict by default

set_predict_request(*, return_std: bool | None | str = '$UNCHANGED$', type_: bool | None | str = '$UNCHANGED$') → SkorchCNNReshaperProba

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.

New 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:

• return_std (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for return_std parameter in predict.

• type (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for type_ parameter in predict.

Returns:

self – The updated object.

Return type:

object

var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd']

Variance type to return from predict by default

class sklearn_utilities.torch.skorch.proba.SkorchReshaperProba(estimator: TEstimator, m_type: Literal['mean', 'median', 'nanmean', 'nanmedian'] = 'mean', var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd'] = 'std')

Bases: SkorchReshaperProbaMixin[TEstimator], SkorchReshaper[TEstimator], Generic[TEstimator]

estimator: TEstimator

m_type: Literal['mean', 'median', 'nanmean', 'nanmedian']

M-statistics type to return from predict by default

set_predict_request(*, return_std: bool | None | str = '$UNCHANGED$', type_: bool | None | str = '$UNCHANGED$') → SkorchReshaperProba

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.

New 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:

• return_std (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for return_std parameter in predict.

• type (str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED) – Metadata routing for type_ parameter in predict.

Returns:

self – The updated object.

Return type:

object

var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd']

Variance type to return from predict by default

class sklearn_utilities.torch.skorch.proba.SkorchReshaperProbaMixin(estimator: TEstimator, *args: Any, m_type: Literal['mean', 'median', 'nanmean', 'nanmedian'] = 'mean', var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd'] = 'std', **kwargs: Any)

Bases: Generic[TEstimator]

skorch wrapper mixin that converts quantile predictions to mean and std.

estimator: TEstimator

m_type: Literal['mean', 'median', 'nanmean', 'nanmedian']

M-statistics type to return from predict by default

predict(X: TX, *, return_std: bool = False, type_: Literal['mean', 'median', 'nanmean', 'nanmedian', 'var', 'std', 'ptp', 'nanvar', 'nanstd'] | tuple[Literal['mean', 'median', 'nanmean', 'nanmedian'], Literal['var', 'std', 'ptp', 'nanvar', 'nanstd']] | None = None, **predict_params: Any) → TY | tuple[TY, TY]

var_type: Literal['var', 'std', 'ptp', 'nanvar', 'nanstd']

Variance type to return from predict by default

class sklearn_utilities.torch.skorch.proba.XSigmoidErrors(*args, **kwargs)

Bases: Module

XSigmoid errors. Loss = x * (2 * sigmoid(x) - 1)

See also

https

//github.com/tuantle/regression-losses-pytorch

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class sklearn_utilities.torch.skorch.proba.XTanhErrors(*args, **kwargs)

Bases: Module

XTanh errors. Loss = x * tanh(x)

See also

https

//github.com/tuantle/regression-losses-pytorch

forward(y_pred: Tensor, y_true: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

sklearn_utilities.torch.skorch.reshaper module

class sklearn_utilities.torch.skorch.reshaper.SkorchCNNReshaper(estimator: TEstimator, *, window_size: int | None)

Bases: EstimatorWrapperBase[TEstimator], Generic[TEstimator]

skorch wrapper that reshapes tabular data for CNNs using sliding windows.

estimator: TEstimator

fit(X: TX, y: TY, **fit_params: Any) → Self

predict(X: TX, **predict_params: Any) → TY

class sklearn_utilities.torch.skorch.reshaper.SkorchReshaper(estimator: TEstimator)

Bases: EstimatorWrapperBase[TEstimator], Generic[TEstimator]

skorch wrapper that reshapes tabular data for NNs.

estimator: TEstimator

fit(X: TX, y: TY, **fit_params: Any) → Self

predict(X: TX, **predict_params: Any) → TY