uqmodels.modelization.DL_estimator package

Submodules

uqmodels.modelization.DL_estimator.Lstm_WS_ED module

class uqmodels.modelization.DL_estimator.Lstm_WS_ED.LSTM_ED_Generator(X, y, metamodel, batch_min=64, shuffle=True, train=True)

Bases: Sequence

load(idx)

on_epoch_end()

Method called at the end of every epoch.

class uqmodels.modelization.DL_estimator.Lstm_WS_ED.Lstm(model_parameters, model_specifications, architecture_parameters, training_parameters={})

Bases: UQEstimator

Build_generator(X, y, batch_min=32, shuffle=True, train=True)

delete()

factory(X, y, mask=None, cut_param=None, fit_rescale=True, causality_remove=None, redundancy=None, only_fit_scaler=False)

Feature factory Reshape and redundundization (Moving window embedding representation)

Parameters:

• X (_type_) – X list contains (X_ctx, X_seq)

• y (_type_) – Raw Y

• mask (_type_, optional) – Mask of non-data

• cut_param (_type_, optional) – cut paramaters on y distrubution

Returns:

model Inputs, Targets and mask.

Return type:

(Inputs,targets,mask)

fit(Inputs, Targets=None, validation_data=None, epochs=None, steps_per_epoch=None, b_s=None, l_r=None, list_loss=None, param_loss=None, shuffle=True, sample_weight=None, verbose=None, metrics=None, callbacks=None, generator=None, validation_freq=1, **kwargs)

Fit UQestimator using training data. :param X: train features :param y: train targets/observations

get_params()

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:

params – Parameter names mapped to their values.

Return type:

dict

load(path, name=None)

model_initializer()

modify_dropout(dp)

Method to modify dp : have to be extend as API modification to model features.

Parameters:

dp (_type_) – ddropout rate.

plot_metrics(name_loss='val_loss')

Plot metrics values recovers form tensorflow metrics callback

Parameters:

name_loss (str, optional) – metrics to visualisze.

predict(Inputs, n_ech=6, mask_h=0, mask_m=[0], generator=None, **kwargs)

Predict procedure Maybe partly redundant to NN fiting procedure : Refactoring needed :param Inputs: Model inputs :type Inputs: List of NN features :param n_ech: Number of MC-DP inferences. :type n_ech: int, optional

Returns:

Meta-model output tuple of (Prediction,Var_A,Var_E)

Return type:

output

reset()

save(path, name=None)

set_fit_request(*, Inputs: bool | None | str = '$UNCHANGED$', Targets: bool | None | str = '$UNCHANGED$', b_s: bool | None | str = '$UNCHANGED$', callbacks: bool | None | str = '$UNCHANGED$', epochs: bool | None | str = '$UNCHANGED$', generator: bool | None | str = '$UNCHANGED$', l_r: bool | None | str = '$UNCHANGED$', list_loss: bool | None | str = '$UNCHANGED$', metrics: bool | None | str = '$UNCHANGED$', param_loss: bool | None | str = '$UNCHANGED$', sample_weight: bool | None | str = '$UNCHANGED$', shuffle: bool | None | str = '$UNCHANGED$', steps_per_epoch: bool | None | str = '$UNCHANGED$', validation_data: bool | None | str = '$UNCHANGED$', validation_freq: bool | None | str = '$UNCHANGED$', verbose: bool | None | str = '$UNCHANGED$') → Lstm

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.Pipeline. Otherwise it has no effect.

Parameters:

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

Returns:

self – The updated object.

Return type:

object

set_predict_request(*, Inputs: bool | None | str = '$UNCHANGED$', generator: bool | None | str = '$UNCHANGED$', mask_h: bool | None | str = '$UNCHANGED$', mask_m: bool | None | str = '$UNCHANGED$', n_ech: bool | None | str = '$UNCHANGED$') → Lstm

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.Pipeline. Otherwise it has no effect.

Parameters:

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

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

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

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

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

Returns:

self – The updated object.

Return type:

object

uqmodels.modelization.DL_estimator.Lstm_WS_ED.double_roll(y, size_window_past, size_window_futur, size_subseq_dec, padding_past, padding_futur, only_futur=False)

uqmodels.modelization.DL_estimator.baseline_models module

uqmodels.modelization.DL_estimator.baseline_models.cnn_mlp(dim_dyn, dim_target, size_window=40, n_windows=10, step=1, dim_chan=1, dim_z=50, type_output='MC_Dropout', dp=0.08, name='')

CNN processing with timed distributed MLP

[ | | | ] *10

mlp mlp mlp val val val

Parameters:

• dim_dyn (_type_) – _description_

• dim_target (_type_) – _description_

• size_window (int, optional) – _description_. Defaults to 40.

• n_windows (int, optional) – _description_. Defaults to 10.

• step (int, optional) – _description_. Defaults to 1.

• dim_chan (int, optional) – _description_. Defaults to 1.

• dim_z (int, optional) – _description_. Defaults to 50.

• name (str, optional) – _description_. Defaults to “”.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.data_embedding module

class uqmodels.modelization.DL_estimator.data_embedding.Conv1D(*args, **kwargs)

Bases: Conv1D

class uqmodels.modelization.DL_estimator.data_embedding.Conv2D(*args, **kwargs)

Bases: Conv2D

class uqmodels.modelization.DL_estimator.data_embedding.DataEmbedding_ITS(*args, **kwargs)

Bases: Layer

call(inputs, x_mark=None, **kwargs)

_summary_

Parameters:

• inputs (_type_) – _description_

• x_mark (_type_, optional) – _description_. Defaults to None.

Returns:

_description_

Return type:

_type_

class uqmodels.modelization.DL_estimator.data_embedding.Data_embedding_TS(*args, **kwargs)

Bases: Layer

call(Z_enc, Y_past, extension)

_summary_

Parameters:

• Z_enc (_type_) – _description_

• Y_past (_type_) – _description_

• Y_futur (_type_) – _description_

• mode (str, optional) – _description_. Defaults to “”.

Returns:

_description_

Return type:

_type_

class uqmodels.modelization.DL_estimator.data_embedding.Dropout(*args, **kwargs)

Bases: Dropout

class uqmodels.modelization.DL_estimator.data_embedding.Factice_Time_Extension(*args, **kwargs)

Bases: Layer

call(inputs, **kwargs)

_summary_

Parameters:

inputs (_type_) – _description_

Returns:

_description_

Return type:

_type_

get_config()

Returns the config of the layer.

A layer config is a Python dictionary (serializable) containing the configuration of a layer. The same layer can be reinstantiated later (without its trained weights) from this configuration.

The config of a layer does not include connectivity information, nor the layer class name. These are handled by Network (one layer of abstraction above).

Note that get_config() does not guarantee to return a fresh copy of dict every time it is called. The callers should make a copy of the returned dict if they want to modify it.

Returns:

Python dictionary.

class uqmodels.modelization.DL_estimator.data_embedding.FixedEmbedding(*args, **kwargs)

Bases: Layer

call(inputs, **kargs)

_summary_

Parameters:

inputs (_type_) – _description_

Returns:

_description_

Return type:

_type_

class uqmodels.modelization.DL_estimator.data_embedding.Mouving_Window_Embedding(*args, **kwargs)

Bases: Layer

call(inputs, mode='encoder')

_summary_

Parameters:

• inputs (_type_) – _description_

• mode (str, optional) – _description_. Defaults to “encoder”.

Returns:

_description_

Return type:

_type_

class uqmodels.modelization.DL_estimator.data_embedding.Mouving_Windows_Embedding(*args, **kwargs)

Bases: Layer

call(inputs, mode='encoder')

_summary_

Parameters:

• inputs (_type_) – _description_

• mode (str, optional) – _description_. Defaults to “encoder”.

Returns:

_description_

Return type:

_type_

class uqmodels.modelization.DL_estimator.data_embedding.Mouving_conv_Embedding(*args, **kwargs)

Bases: Layer

call(inputs)

_summary_

Parameters:

• inputs (_type_) – _description_

• mode (str, optional) – _description_. Defaults to “encoder”.

Returns:

_description_

Return type:

_type_

classmethod from_config(config)

Creates a layer from its config.

This method is the reverse of get_config, capable of instantiating the same layer from the config dictionary. It does not handle layer connectivity (handled by Network), nor weights (handled by set_weights).

Parameters:

config – A Python dictionary, typically the output of get_config.

Returns:

A layer instance.

get_config()

Returns the config of the layer.

A layer config is a Python dictionary (serializable) containing the configuration of a layer. The same layer can be reinstantiated later (without its trained weights) from this configuration.

The config of a layer does not include connectivity information, nor the layer class name. These are handled by Network (one layer of abstraction above).

Note that get_config() does not guarantee to return a fresh copy of dict every time it is called. The callers should make a copy of the returned dict if they want to modify it.

Returns:

Python dictionary.

class uqmodels.modelization.DL_estimator.data_embedding.PositionalEmbedding(*args, **kwargs)

Bases: Layer

call(inputs, **kwargs)

_summary_

Parameters:

inputs (_type_) – _description_

Returns:

_description_

Return type:

_type_

class uqmodels.modelization.DL_estimator.data_embedding.TemporalEmbedding(*args, **kwargs)

Bases: Layer

call(inputs, **kargs)

_summary_

Parameters:

inputs (_type_) – _description_

Returns:

_description_

Return type:

_type_

class uqmodels.modelization.DL_estimator.data_embedding.ValuesEmbedding(*args, **kwargs)

Bases: Layer

call(inputs, **kwargs)

_summary_

Parameters:

inputs (_type_) – _description_

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.loss module

uqmodels.modelization.DL_estimator.loss.BNN_loss_edl(true, pred, alpha=0.95)

uqmodels.modelization.DL_estimator.loss.BNN_loss_gaussian(true, pred, alpha=0.95)

uqmodels.modelization.DL_estimator.loss.BNN_metric_edl(true, pred)

uqmodels.modelization.DL_estimator.loss.BNN_metric_gaussian(true, pred)

uqmodels.modelization.DL_estimator.loss.build_BNN_loss(alpha=0.95, metric=False, type_output='MC_Dropout')

uqmodels.modelization.DL_estimator.loss.build_CCE_loss()

uqmodels.modelization.DL_estimator.loss.build_EDL_loss(coeff_reg=0.95, coeff_var_pen=1)

uqmodels.modelization.DL_estimator.loss.build_MSE_loss(split=2, metric=False, var=1)

uqmodels.modelization.DL_estimator.loss.default_callbacks(min_delta=0.0001, earlystop_patience=60, reducelr_patience=30, reducelr_factor=0.3, reduce_lr_min_lr=1e-06, verbose=0)

uqmodels.modelization.DL_estimator.lstm_ed module

class uqmodels.modelization.DL_estimator.lstm_ed.Lstm_ED_UQ(model_parameters, factory_parameters={}, training_parameters={}, type_output=None, rescale=False, n_ech=5, train_ratio=0.9, name='', random_state=None)

Bases: NN_UQ

Build_generator(X, y, batch=32, shuffle=True, train=True)

factory(X, y, mask=None, only_fit_scaler=False, **kwarg)

set_fit_request(*, Inputs: bool | None | str = '$UNCHANGED$', Targets: bool | None | str = '$UNCHANGED$', test: bool | None | str = '$UNCHANGED$', train: bool | None | str = '$UNCHANGED$', training_parameters: bool | None | str = '$UNCHANGED$', verbose: bool | None | str = '$UNCHANGED$') → Lstm_ED_UQ

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.Pipeline. Otherwise it has no effect.

Parameters:

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

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

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

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

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

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

Returns:

self – The updated object.

Return type:

object

set_predict_request(*, generator: bool | None | str = '$UNCHANGED$', type_output: bool | None | str = '$UNCHANGED$') → Lstm_ED_UQ

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.Pipeline. Otherwise it has no effect.

Parameters:

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

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

Returns:

self – The updated object.

Return type:

object

uqmodels.modelization.DL_estimator.lstm_ed.build_lstm_stacked(size_window=20, n_windows=5, step=1, dim_target=3, dim_chan=1, dim_horizon=5, dim_ctx=18, dim_z=200, dp=0.05, dp_rec=0.03, type_output=None, num_lstm_enc=1, num_lstm_dec=1, k_reg=(1e-05, 1e-05), layers_enc=[75, 150, 75], layers_dec=[150, 75], list_strides=[2, 1], list_kernels=None, list_filters=None, with_ctx_input=True, with_convolution=True, dim_dyn=None, random_state=None, **kwarg)

Builder for LSTM ED UQ with convolutif preprocessing for lag values

Parameters:

• size_window (int, optional) – Size of window for lag values. Defaults to 10.

• n_windows (int, optional) – Number of window in past. Defaults to 5.

• step (int, optional) – step between windows. Defaults to 1.

• dim_target (int, optional) – dimension of TS. Defaults to 1.

• dim_chan (int, optional) – Number of channel of TS. Defaults to 1.

• dim_horizon (int, optional) – futur_horizon to predict. Defaults to 3.

• dim_ctx (int, optional) – Number of ctx_features. Defaults to 20.

• dim_z (int, optional) – Size of latent sapce. Defaults to 100.

• layers_enc (list, optional) – size of MLP preprocessing

• [150]. ((after concatenation of past values embeding + ctx) Defaults to)

• layers_dec (list, optional) – size of MLP interpretor. Defaults to 2.

• dp (float, optional) – dropout. Defaults to 0.05.

• dp_rec (float, optional) – transformer dropout. Defaults to 0.1.

• k_reg (tuple, optional) – _description_. Defaults to (0.00001, 0.00001).

• with_positional_embedding (bool, optional) – _description_. Defaults to False.

• with_ctx_input (bool, optional) – Expect ctx features in addition to lag. Defaults to True.

• with_convolution (bool, optional) – use convolution rather than

• True. (whole lag values in the windows. Defaults to)

• type_output (_type_, optional) – mode of UQ (see NN_UQ). Defaults to None.

• random_state (bool) – handle experimental random using seed.

Returns:

multi-step forecaster with UQ

Return type:

transformer

uqmodels.modelization.DL_estimator.lstm_ed.get_params_dict(dim_ctx, dim_dyn, dim_target, size_window=20, n_windows=5, dim_horizon=5, step=1, dim_chan=1, dp=0.05, dp_rec=0.05, dim_z=50, k_reg=(1e-06, 1e-06), num_lstm_enc=1, num_lstm_dec=1, layers_enc=[150, 75], layers_dec=[200, 125, 75], list_strides=[2, 1, 1, 1], list_filters=[128, 128, 128], list_kernels=None, with_convolution=True, with_ctx_input=True, n_ech=3, type_output='MC_Dropout')

uqmodels.modelization.DL_estimator.metalayers module

class uqmodels.modelization.DL_estimator.metalayers.Add_query_to_Z_Processing_with_state(*args, **kwargs)

Bases: Layer

call(ZProcessing, Query)

_summary_

Parameters:

• ZProcessing (_type_) – _description_

• Query (_type_) – _description_

Returns:

_description_

Return type:

_type_

compute_output_shape(input_shape)

Computes the output shape of the layer.

This method will cause the layer’s state to be built, if that has not happened before. This requires that the layer will later be used with inputs that match the input shape provided here.

Parameters:

input_shape – Shape tuple (tuple of integers) or tf.TensorShape, or structure of shape tuples / tf.TensorShape instances (one per output tensor of the layer). Shape tuples can include None for free dimensions, instead of an integer.

Returns:

A tf.TensorShape instance or structure of tf.TensorShape instances.

class uqmodels.modelization.DL_estimator.metalayers.Double_Moving_slice_layer(*args, **kwargs)

Bases: Layer

Layer that apply double moving_slice_map

Parameters:

Layer (_type_) – _description_

build(input_shape)

Creates the variables of the layer (for subclass implementers).

This is a method that implementers of subclasses of Layer or Model can override if they need a state-creation step in-between layer instantiation and layer call. It is invoked automatically before the first execution of call().

This is typically used to create the weights of Layer subclasses (at the discretion of the subclass implementer).

Parameters:

input_shape – Instance of TensorShape, or list of instances of TensorShape if the layer expects a list of inputs (one instance per input).

call(input_data)

_summary_

Parameters:

input_data (_type_) – _description_

Returns:

_description_

Return type:

_type_

compute_output_shape(input_shape)

Computes the output shape of the layer.

This method will cause the layer’s state to be built, if that has not happened before. This requires that the layer will later be used with inputs that match the input shape provided here.

Parameters:

input_shape – Shape tuple (tuple of integers) or tf.TensorShape, or structure of shape tuples / tf.TensorShape instances (one per output tensor of the layer). Shape tuples can include None for free dimensions, instead of an integer.

Returns:

A tf.TensorShape instance or structure of tf.TensorShape instances.

class uqmodels.modelization.DL_estimator.metalayers.EDLProcessing(*args, **kwargs)

Bases: Layer

call(x)

Apply EDLProcessing

Parameters:

x (_type_) – input

Returns:

_description_

Return type:

_type_

compute_output_shape(input_shape)

Computes the output shape of the layer.

This method will cause the layer’s state to be built, if that has not happened before. This requires that the layer will later be used with inputs that match the input shape provided here.

Parameters:

input_shape – Shape tuple (tuple of integers) or tf.TensorShape, or structure of shape tuples / tf.TensorShape instances (one per output tensor of the layer). Shape tuples can include None for free dimensions, instead of an integer.

Returns:

A tf.TensorShape instance or structure of tf.TensorShape instances.

get_config()

Returns the config of the layer.

A layer config is a Python dictionary (serializable) containing the configuration of a layer. The same layer can be reinstantiated later (without its trained weights) from this configuration.

The config of a layer does not include connectivity information, nor the layer class name. These are handled by Network (one layer of abstraction above).

Note that get_config() does not guarantee to return a fresh copy of dict every time it is called. The callers should make a copy of the returned dict if they want to modify it.

Returns:

Python dictionary.

class uqmodels.modelization.DL_estimator.metalayers.LSTMCellMidsize(*args, **kwargs)

Bases: LSTMCell

Hack to take into accout state in cell : size in dim_z*2 | size out dim_z

Parameters:

LSTMCell (_type_) – _description_

build(input_shape)

Creates the variables of the layer (for subclass implementers).

This is a method that implementers of subclasses of Layer or Model can override if they need a state-creation step in-between layer instantiation and layer call. It is invoked automatically before the first execution of call().

This is typically used to create the weights of Layer subclasses (at the discretion of the subclass implementer).

Parameters:

input_shape – Instance of TensorShape, or list of instances of TensorShape if the layer expects a list of inputs (one instance per input).

class uqmodels.modelization.DL_estimator.metalayers.LSTMCellReturnCellState(*args, **kwargs)

Bases: LSTMCell

Layer LSTM returning output and state jointly

Parameters:

LSTMCell (_type_) – _description_

call(inputs, states, training=None)

_summary_

Parameters:

• inputs (_type_) – _description_

• states (_type_) – _description_

• training (_type_, optional) – _description_. Defaults to None.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.metalayers.LSTM_DProcessing(n_step, dim_z, flag_mc, dp=0.05, dp_r=0.02, l1_l2_reg=(1e-07, 1e-07), random_state=None)

Decoder Processing as block : aim to make temporal projection (Usefull to hold query information) Input (batch,n_step,dim_z*2) output (batch,n_step,dim_z) with Zdecoding latent space)

Parameters:

• n_step (_type_) – _description_

• dim_z (_type_) – _description_

• flag_mc (_type_) – _description_

• dp (float, optional) – _description_. Defaults to 0.05.

• dp_r (float, optional) – _description_. Defaults to 0.02.

• l1_l2_reg (tuple, optional) – _description_. Defaults to (0.0000001, 0.0000001).

• random_state (bool) – handle experimental random using seed.

Returns:

Keras model as LSTM Decoder block

Return type:

Model

uqmodels.modelization.DL_estimator.metalayers.LSTM_EProcessing(n_step, dim_in, dim_z, flag_mc, dp=0.05, dp_r=0.02, l1_l2_reg=(1e-07, 1e-07), random_state=None)

Encoder Processing as block aim to capture dynamic Input (batch,n_step,dim_in) output (batch,n_step,dim_z*2) with Z_h and Z_state concatenate)

Parameters:

• n_step (_type_) – _description_

• dim_in (_type_) – _description_

• dim_z (_type_) – _description_

• flag_mc (_type_) – _description_

• dp (float, optional) – _description_. Defaults to 0.05.

• dp_r (float, optional) – _description_. Defaults to 0.02.

• l1_l2_reg (tuple, optional) – _description_. Defaults to (0.0000001, 0.0000001).

• random_state (bool) – handle experimental random using seed.

Returns:

Keras model as LSTM Encoder block

Return type:

Model

class uqmodels.modelization.DL_estimator.metalayers.Moving_slice_layer(*args, **kwargs)

Bases: Layer

Layer that apply moving_slice_map

build(input_shape)

Creates the variables of the layer (for subclass implementers).

This is a method that implementers of subclasses of Layer or Model can override if they need a state-creation step in-between layer instantiation and layer call. It is invoked automatically before the first execution of call().

This is typically used to create the weights of Layer subclasses (at the discretion of the subclass implementer).

Parameters:

input_shape – Instance of TensorShape, or list of instances of TensorShape if the layer expects a list of inputs (one instance per input).

call(input_data)

Apply moving_slice_map to generate window sequnce

Parameters:

input_data (_type_) – _description_

Returns:

_description_

Return type:

_type_

compute_output_shape(input_shape)

Computes the output shape of the layer.

This method will cause the layer’s state to be built, if that has not happened before. This requires that the layer will later be used with inputs that match the input shape provided here.

Parameters:

input_shape – Shape tuple (tuple of integers) or tf.TensorShape, or structure of shape tuples / tf.TensorShape instances (one per output tensor of the layer). Shape tuples can include None for free dimensions, instead of an integer.

Returns:

A tf.TensorShape instance or structure of tf.TensorShape instances.

class uqmodels.modelization.DL_estimator.metalayers.ProbabilisticProcessing(*args, **kwargs)

Bases: Layer

_summary_

Parameters:

Layer (_type_) – _description_

call(x)

Apply ProbabilisticProcessing to x

Parameters:

x (_type_) – _description_

Returns:

_description_

Return type:

_type_

compute_output_shape(input_shape)

Computes the output shape of the layer.

This method will cause the layer’s state to be built, if that has not happened before. This requires that the layer will later be used with inputs that match the input shape provided here.

Parameters:

input_shape – Shape tuple (tuple of integers) or tf.TensorShape, or structure of shape tuples / tf.TensorShape instances (one per output tensor of the layer). Shape tuples can include None for free dimensions, instead of an integer.

Returns:

A tf.TensorShape instance or structure of tf.TensorShape instances.

get_config()

Returns the config of the layer.

A layer config is a Python dictionary (serializable) containing the configuration of a layer. The same layer can be reinstantiated later (without its trained weights) from this configuration.

The config of a layer does not include connectivity information, nor the layer class name. These are handled by Network (one layer of abstraction above).

Note that get_config() does not guarantee to return a fresh copy of dict every time it is called. The callers should make a copy of the returned dict if they want to modify it.

Returns:

Python dictionary.

class uqmodels.modelization.DL_estimator.metalayers.RNN_states_in_inputs(*args, **kwargs)

Bases: RNN

RNN class dispatching jointly [H,C]

Parameters:

RNN (_type_) – _description_

call(inputs, mask=None, training=None, initial_state=None, constants=None)

_summary_

Parameters:

• inputs (_type_) – _description_

• mask (_type_, optional) – _description_. Defaults to None.

• training (_type_, optional) – _description_. Defaults to None.

• initial_state (_type_, optional) – _description_. Defaults to None.

• constants (_type_, optional) – _description_. Defaults to None.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.metalayers.Tconv_block_1D(inputs, dim_out, filters=32, kernel=2, strides=2, dp=0.02, flag_mc=False, random_state=None)

uqmodels.modelization.DL_estimator.metalayers.Tconv_block_2D(inputs, dim_out, filters=32, kernel=5, strides=(2, 1), dp=0.02, flag_mc=False, random_state=None)

uqmodels.modelization.DL_estimator.metalayers.cnn_dec(size_subseq_dec, dim_out, dim_chan=1, type_output=None, k1=10, min_logvar=-6, dim_z=100, dp=0.01, random_state=None, **kwarg)

Warning depreciated CNN_dec implementation

Parameters:

• size_subseq_dec (_type_) – _description_

• dim_out (_type_) – _description_

• dim_chan (int, optional) – _description_. Defaults to 1.

• type_output (_type_, optional) – _description_. Defaults to None.

• k1 (int, optional) – _description_. Defaults to 10.

• min_logvar (int, optional) – _description_. Defaults to -6.

• dim_z (int, optional) – _description_. Defaults to 100.

• dp (float, optional) – _description_. Defaults to 0.01.

• random_state (_type_, optional) – _description_. Defaults to None.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.metalayers.cnn_dec_1D(size_subseq_dec=10, dim_out=10, dim_z=100, type_output=None, k=32, dp=0.05, random_state=None, **kwarg)

_summary_

Parameters:

• size_subseq_dec (int, optional) – _description_. Defaults to 10.

• dim_out (int, optional) – _description_. Defaults to 10.

• dim_z (int, optional) – _description_. Defaults to 100.

• type_output (_type_, optional) – _description_. Defaults to None.

• k (int, optional) – _description_. Defaults to 32.

• dp (float, optional) – _description_. Defaults to 0.05.

• random_state (bool) – handle experimental random using seed.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.metalayers.cnn_dec_bis(size_subseq_dec, dim_out, dim_chan=1, type_output=None, min_logvar=-6, list_filters=[64, 64], strides=(1, 1), list_kernels=[4, 4], dim_z=200, random_state=None, **kwarg)

uqmodels.modelization.DL_estimator.metalayers.cnn_enc(size_subseq_enc, dim_out, dim_chan=1, k1=10, reduction_x1=8, reduction_x2=1, dim_z=100, dp=0.02, random_state=None, **kwarg)

Warning depreciated CNN_enc implementation

Parameters:

• size_subseq_enc (_type_) – _description_

• dim_out (_type_) – _description_

• dim_chan (int, optional) – _description_. Defaults to 1.

• k1 (int, optional) – _description_. Defaults to 10.

• reduction_x1 (int, optional) – _description_. Defaults to 8.

• reduction_x2 (int, optional) – _description_. Defaults to 1.

• dim_z (int, optional) – _description_. Defaults to 100.

• dp (float, optional) – _description_. Defaults to 0.02.

• random_state (_type_, optional) – _description_. Defaults to None.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.metalayers.cnn_enc_1D(size_subseq_enc, dim_out, dim_z, dim_lt, type_output=None, k=32, dp=0.05, random_state=None, **kwarg)

cnn_enc_1D layers

Parameters:

• size_subseq_enc (int, optional) – _description_. Defaults to 100.

• dim_out (int, optional) – _description_. Defaults to 10.

• dim_z (int, optional) – _description_. Defaults to 100.

• dim_lt (int, optional) – _description_. Defaults to 100.

• type_output (_type_, optional) – _description_. Defaults to None.

• k (int, optional) – _description_. Defaults to 32.

• dp (float, optional) – _description_. Defaults to 0.05.

• random_state (bool) – handle experimental random using seed.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.metalayers.cnn_enc_bis(size_subseq_enc=60, dim_target=52, dim_chan=4, list_filters=[64, 64, 32], list_kernels=[(10, 3), 10, 10], list_strides=[(2, 1), (2, 1), (2, 1)], type_output=None, block='2D', dim_z=200, dp=0.02, random_state=None, **kwarg)

Produce a cnn_enn subpart of a deep learning predictor

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.metalayers.conv_block_1D(inputs, dim_chan, filters=32, kernel=2, strides=2, dp=0.02, flag_mc=False, random_state=None)

uqmodels.modelization.DL_estimator.metalayers.conv_block_2D(inputs, dim_chan, filters=32, kernel=5, strides=(2, 1), dp=0.02, flag_mc=False, random_state=None)

uqmodels.modelization.DL_estimator.metalayers.dense2D_enc_dec(size_subseq_enc, size_subseq_dec, dim_in, dim_out, layers_size=[100, 50], dim_z=100, dp=0.05, enc_only=False, type_output=None, random_state=None)

uqmodels.modelization.DL_estimator.metalayers.get_cnn_dec_params(dim_target, size_subseq_dec=1, dim_z=50, random_state=None)

Produce dict params that can instanciate cnn_dec_bloc :param dim_target: dimension of motifs to convolute :type dim_target: _type_ :param size_subseq_dec: length of motifs to convulute :type size_subseq_dec: int, optional :param dim_z: latent dimension :type dim_z: int, optional :param random_state: handle experimental random using seed. :type random_state: bool

uqmodels.modelization.DL_estimator.metalayers.get_cnn_enc_params(dim_target, size_subseq_enc=1, dim_z=50, random_state=None)

Produce dict params that can instanciate cnn_enn_bloc :param dim_target: dimension of motifs to convolute :type dim_target: _type_ :param size_subseq_enc: length of motifs to convulute :type size_subseq_enc: int, optional :param dim_z: latent dimension :type dim_z: int, optional :param random_state: handle experimental random using seed. :type random_state: bool

uqmodels.modelization.DL_estimator.metalayers.mlp(dim_in=10, dim_out=1, layers_size=[100, 50], name='', dp=0.01, with_mc_dp=True, type_output=None, logvar_min=-10, regularizer_W=(1e-05, 1e-05), shape_2D=None, shape_2D_out=None, random_state=None, **kwargs)

Generate a keras MLP model to make preprocessing or head subpart”.

Parameters:

• dim_in (int) – Input dimension, erase by shape_2D if 2D input_size

• dim_out (int or None) – Input dimension, if None take the last layers_size values

• layers_size (list of in, optional) – List of size of layers. Defaults to [100, 50].

• name (str, optional) – Name of model. Defaults to “”.

• dp (float, optional) – Percentage of dropout. Defaults to 0.01.

• type_output (_type_, optional) – Specify Head last layers among

• ['None' – pred,MC_Dropout:(Pred,var),”EDL”:(Pred,mu,alpha,beta) ]

• logvar_min (int, optional) – Cut off for small variance estimations

• regularizer_W (tuple, optional) – Regularisation on Dense layers. Defaults to (0.00001, 0.00001).

• shape_2D (tupple or None, optional) – if intput shape is 2D. Defaults to None.

• shape_2D_out

• random_state (bool) – handle experimental random using seed.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.metalayers.moving_slice_map(inputs, n_step, padding, kick_off=0, depth_slice=1)

Apply Layers on n_step slices of input with padding :param input: Input tensors :type input: TF.tensor :param Layers: Submodels :type Layers: Keras.model :param n_step: N_slide :type n_step: int :param padding: padding :type padding: int

uqmodels.modelization.DL_estimator.metalayers.stack_and_roll_layer(inputs, size_window, size_subseq, padding, name='', format='tf_slice')

Layers that produce a stack rolled layers to produce a batch of subsequence

Parameters:

• inputs (_type_) – layers

• size_window (_type_) – size of subseqeuence

• size_subseq (_type_) – _description_

• padding (_type_) – _description_

• name (str, optional) – _description_. Defaults to “”.

• format (str, optional) – _description_. Defaults to “tf_slice”.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.neural_network_UQ module

uqmodels.modelization.DL_estimator.neural_network_UQ.Deterministic_prediction(Inputs, model, ddof, generator=False, type_output=None)

Prediction (mu,sigma) of Inputs using Deterministic UQ-paragim (Ex : EDL)

Parameters:

• model (tf.model) – neural network

• n_ech (n_draw) – number of dropout drawn

• Inputs (_type_) – Inputs of model

• ddof (_type_) – ddof

• generator (bool, optional) – specify if Inputs is generator or not. Defaults to False.

• type_output – type_output (EDL)

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.neural_network_UQ.Drawn_based_prediction(Inputs, model, n_ech, ddof, generator=False, type_output='MC_Dropout')

Prediction (mu,sigma) of Inputs using Drawn_based UQ-paragim (Ex : MC_dropout)

Parameters:

• model (tf.model) – neural network

• n_ech (n_draw) – number of dropout drawn

• Inputs (_type_) – Inputs of model

• ddof (_type_) – ddof

• generator (bool, optional) – specify if Inputs is generator or not. Defaults to False.

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.neural_network_UQ.Ensemble_based_prediction(Inputs, models, ddof, generator=False, type_output=None)

Prediction (mu,sigma) of Inputs using Ensemble_based UQ-paradign

Parameters:

• model (tf.model) – neural network

• n_ech (n_draw) – number of dropout drawn

• Inputs (_type_) – Inputs of model

• ddof (_type_) – ddof

• generator (bool, optional) – specify if Inputs is generator or not. Defaults to False.

• type_output – type_output (curently useless)

Returns:

_description_

Return type:

_type_

uqmodels.modelization.DL_estimator.neural_network_UQ.Identity_factory(X, y, **kwargs)

class uqmodels.modelization.DL_estimator.neural_network_UQ.NN_UQ(model_initializer, model_parameters, factory_parameters={}, training_parameters={}, type_output=None, rescale=False, n_ech=5, train_ratio=0.9, var_min=1e-06, name='NN', random_state=None)

Bases: UQEstimator

Neural Network UQ

Build_generator(X, y, batch=32, shuffle=True, train=True)

basic_fit(Inputs, Targets, train=None, test=None, epochs=[1000, 1000], b_s=[100, 20], l_r=[0.01, 0.005], sample_w=None, verbose=1, list_loss=['mse'], metrics=None, generator=None, steps_per_epoch=None, shuffle=True, callbacks='default', validation_freq=1, param_loss=None, test_batch_size=None, **kwargs)

basic_predict(Inputs, n_ech=6, type_output='MC_Dropout', generator=None, test_batch_size=None, **kwarg)

build_loss(loss, param_loss=None)

Build loss from str or loss and loss_parameters

Parameters:

• loss (_type_) – _description_

• param_loss (_type_, optional) – _description_. Defaults to None.

Returns:

_description_

Return type:

_type_

build_metrics(metrics)

Build list of metrics from str or metrics.

Parameters:

metrics (_type_) – _description_

compile(step=0, optimizer=None, loss=None, metrics=None, **kwarg)

dataset_generator(Inputs, Targets, validation_data, batch_size, shuffle, generator, test_batch_size=None)

Hold case with or without data generator

Parameters:

• Inputs (_type_) – _description_

• Targets (_type_) – _description_

• validation_data (_type_) – _description_

• batch (_type_) – _description_

• shuffle (_type_) – _description_

• generator (_type_) – _description_

Returns:

_description_

Return type:

_type_

factory(X, y, mask=None, cut_param=None, only_fit_scaler=False)

fit(Inputs, Targets, train=None, test=None, training_parameters=None, verbose=None, **kwargs)

Fit UQestimator using training data. :param X: train features :param y: train targets/observations

init_neural_network()

apply model_initializer function with model_parameters and store in self.model

load(path, name=None)

modify_dropout(dp)

predict(X, type_output=None, generator=None, **kwargs)

Compute prediction (or provide None) and UQ-measure :param X: features

Returns:

pred, UQ_measure

reset()

save(path=None, name=None)

set_fit_request(*, Inputs: bool | None | str = '$UNCHANGED$', Targets: bool | None | str = '$UNCHANGED$', test: bool | None | str = '$UNCHANGED$', train: bool | None | str = '$UNCHANGED$', training_parameters: bool | None | str = '$UNCHANGED$', verbose: bool | None | str = '$UNCHANGED$') → NN_UQ

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.Pipeline. Otherwise it has no effect.

Parameters:

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

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

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

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

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

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

Returns:

self – The updated object.

Return type:

object

set_predict_request(*, generator: bool | None | str = '$UNCHANGED$', type_output: bool | None | str = '$UNCHANGED$') → NN_UQ

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.Pipeline. Otherwise it has no effect.

Parameters:

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

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

Returns:

self – The updated object.

Return type:

object

class uqmodels.modelization.DL_estimator.neural_network_UQ.default_Generator(X, y, metamodel, batch=64, shuffle=True, train=True, random_state=None)

Bases: Sequence

load(idx)

on_epoch_end()

Method called at the end of every epoch.

uqmodels.modelization.DL_estimator.neural_network_UQ.generate_K_fold_removing_index(n_model, k_fold, train, data_drop, random_state=None)

Generate liste of idx to remove for k_fold deep ensemble procedure

Parameters:

• n_model (_type_) – Number of models

• k_fold (_type_) – Number of fold

• train (_type_) – train_flag_idx

• data_drop (_type_) – % of data drop

• random_state – handle experimental random using seed

Returns:

list_sampletoremove idx of sample to remove of train for each submodel

Return type:

_type_

uqmodels.modelization.DL_estimator.neural_network_UQ.generate_train_test(len_, train_ratio=0.92, last_val=True, random_state=None)

uqmodels.modelization.DL_estimator.neural_network_UQ.get_params_dict(dim_in, dim_out=1, layers_size=[200, 150, 100], regularizer_=(0.0001, 0.0001), dp=None, name='MLP_UQ', type_output='MC_Dropout', **kwargs)

uqmodels.modelization.DL_estimator.neural_network_UQ.get_training_parameters(epochs=[100, 100], b_s=[64, 32], l_r=[0.005, 0.001], list_loss=None, metrics=None, param_loss=None, type_output=None, generator=False, shuffle=True, verbose=1, sample_w=None, callbacks='default', **kwargs)

uqmodels.modelization.DL_estimator.transformer_ed module

class uqmodels.modelization.DL_estimator.transformer_ed.Dense(*args, **kwargs)

Bases: Dense

class uqmodels.modelization.DL_estimator.transformer_ed.Dropout(*args, **kwargs)

Bases: Dropout

class uqmodels.modelization.DL_estimator.transformer_ed.LayerNormalization(*args, **kwargs)

Bases: LayerNormalization

class uqmodels.modelization.DL_estimator.transformer_ed.MultiHeadAttention(*args, **kwargs)

Bases: MultiHeadAttention

class uqmodels.modelization.DL_estimator.transformer_ed.TransformerDecoder(*args, **kwargs)

Bases: Layer

Transformer Encoder Layer from https://keras.io/examples/audio/transformer_asr/

call(enc_out, target, training=None)

_summary_

Parameters:

• enc_out (_type_) – _description_

• target (_type_) – _description_

Returns:

_description_

Return type:

_type_

causal_attention_mask(batch_size, n_dest, n_src, dim_horizon, dtype)

Masks the upper half of the dot product matrix in self attention.

This prevents flow of information from future tokens to current token. 1’s in the lower triangle, counting from the lower right corner.

classmethod from_config(config)

Creates a layer from its config.

This method is the reverse of get_config, capable of instantiating the same layer from the config dictionary. It does not handle layer connectivity (handled by Network), nor weights (handled by set_weights).

Parameters:

config – A Python dictionary, typically the output of get_config.

Returns:

A layer instance.

get_config()

Returns the config of the layer.

A layer config is a Python dictionary (serializable) containing the configuration of a layer. The same layer can be reinstantiated later (without its trained weights) from this configuration.

The config of a layer does not include connectivity information, nor the layer class name. These are handled by Network (one layer of abstraction above).

Note that get_config() does not guarantee to return a fresh copy of dict every time it is called. The callers should make a copy of the returned dict if they want to modify it.

Returns:

Python dictionary.

class uqmodels.modelization.DL_estimator.transformer_ed.TransformerEncoder(*args, **kwargs)

Bases: Layer

Transformer Encoder Layer from https://keras.io/examples/audio/transformer_asr/

call(inputs, training=None)

_summary_

Parameters:

• inputs (_type_) – _description_

• training (_type_) – _description_

Returns:

_description_

Return type:

_type_

classmethod from_config(config)

Creates a layer from its config.

This method is the reverse of get_config, capable of instantiating the same layer from the config dictionary. It does not handle layer connectivity (handled by Network), nor weights (handled by set_weights).

Parameters:

config – A Python dictionary, typically the output of get_config.

Returns:

A layer instance.

get_config()

Returns the config of the layer.

A layer config is a Python dictionary (serializable) containing the configuration of a layer. The same layer can be reinstantiated later (without its trained weights) from this configuration.

The config of a layer does not include connectivity information, nor the layer class name. These are handled by Network (one layer of abstraction above).

Note that get_config() does not guarantee to return a fresh copy of dict every time it is called. The callers should make a copy of the returned dict if they want to modify it.

Returns:

Python dictionary.

class uqmodels.modelization.DL_estimator.transformer_ed.Transformer_ED_UQ(model_parameters, factory_parameters={'factory_lag_lt': 0, 'factory_lag_st': 0}, training_parameters={}, type_output=None, rescale=False, n_ech=5, train_ratio=0.9, name='Lstm_stacked', random_state=None)

Bases: NN_UQ

Transformer_ED for forecasting with UQ : see build_transformer to check model parameters

Build_generator(X, y, batch=32, shuffle=True, train=True)

factory(X, y, mask=None, only_fit_scaler=False, **kwarg)

set_fit_request(*, Inputs: bool | None | str = '$UNCHANGED$', Targets: bool | None | str = '$UNCHANGED$', test: bool | None | str = '$UNCHANGED$', train: bool | None | str = '$UNCHANGED$', training_parameters: bool | None | str = '$UNCHANGED$', verbose: bool | None | str = '$UNCHANGED$') → Transformer_ED_UQ

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.Pipeline. Otherwise it has no effect.

Parameters:

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

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

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

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

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

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

Returns:

self – The updated object.

Return type:

object

set_predict_request(*, generator: bool | None | str = '$UNCHANGED$', type_output: bool | None | str = '$UNCHANGED$') → Transformer_ED_UQ

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.Pipeline. Otherwise it has no effect.

Parameters:

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

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

Returns:

self – The updated object.

Return type:

object

uqmodels.modelization.DL_estimator.transformer_ed.build_transformer(size_window=10, n_windows=5, step=1, dim_target=1, dim_chan=1, dim_horizon=3, dim_ctx=20, dim_z=100, num_heads=2, num_feed_forward=128, num_layers_enc=3, num_layers_dec=2, layers_enc=[150], layers_dec=[150, 75], dp=0.05, dp_rec=0.03, k_reg=(1e-05, 1e-05), list_strides=[2, 1], list_filters=None, list_kernels=None, dim_dyn=None, with_positional_embedding=False, with_ctx_input=True, with_convolution=True, type_output=None, random_state=None, **kwargs)

Builder for Transformer ED with convolutive preprocessing

Parameters:

• size_window (int, optional) – Size of window for lag values. Defaults to 10.

• n_windows (int, optional) – Number of window in past. Defaults to 5.

• step (int, optional) – step between windows. Defaults to 1.

• dim_target (int, optional) – dimension of TS. Defaults to 1.

• dim_chan (int, optional) – Number of channel of TS. Defaults to 1.

• dim_horizon (int, optional) – futur_horizon to predict. Defaults to 3.

• dim_ctx (int, optional) – Number of ctx_features. Defaults to 20.

• dim_z (int, optional) – Size of latent sapce. Defaults to 100.

• num_heads (int, optional) – num of heads transformer. Defaults to 2.

• num_feed_forward (int, optional) – feed_forward transfomer dimension. Defaults to 128.

• num_layers_enc (int, optional) – num of transformer enc block

• 3. ((after concatenation of past values embeding + ctx) . Defaults to)

• num_layers_dec (int, optional) – num of transformer dec block Defaults to 2.

• layers_enc (list, optional) – size of MLP preprocessing

• [150]. ((after concatenation of past values embeding + ctx) Defaults to)

• layers_dec (list, optional) – size of MLP interpretor. Defaults to 2.

• dp (float, optional) – dropout. Defaults to 0.05.

• dp_t (float, optional) – transformer dropout. Defaults to 0.1.

• k_reg (tuple, optional) – _description_. Defaults to (0.00001, 0.00001).

• dim_dyn (int, None) – size of dyn inputs, if None consider dim_dyn have same size than dim target

• with_positional_embedding (bool, optional) – _description_. Defaults to False.

• with_ctx_input (bool, optional) – Expect ctx features in addition to lag. Defaults to True.

• with_convolution (bool, optional) – use convolution rather than

• True. (whole lag values in the windows. Defaults to)

• type_output (_type_, optional) – mode of UQ (see NN_UQ). Defaults to None.

• random_state (bool) – handle experimental random using seed.

Returns:

multi-step forecaster with UQ

Return type:

transformer

uqmodels.modelization.DL_estimator.transformer_ed.get_params_dict(dim_ctx, dim_dyn, dim_target, dim_chan=1, size_window=20, n_windows=5, dim_horizon=5, dim_z=50, dp=0.05, dp_rec=0.02, num_heads=2, num_feed_forward=128, num_layers_enc=3, num_layers_dec=2, layers_enc=[75, 150, 75], layers_dec=[200, 125, 75], list_strides=[2, 1, 1, 1], list_filters=[128, 128, 128], list_kernels=None, with_convolution=True, with_ctx_input=True, n_ech=3, type_output='MC_Dropout', random_state=None)

uqmodels.modelization.DL_estimator.utils module

class uqmodels.modelization.DL_estimator.utils.Folder_Generator(X, y, metamodel, batch=64, shuffle=True, train=True, random_state=None)

Bases: Sequence

load(idx)

load seq of data locate at [idx*self.batch-past_horizon, idx*self.batch+self.futur_horizon]

on_epoch_end()

Method called at the end of every epoch.

class uqmodels.modelization.DL_estimator.utils.Generator(X, y, batch_min=64, shuffle=True, random_state=None)

Bases: object

load(idx)

on_epoch_end()

uqmodels.modelization.DL_estimator.utils.find_conv_kernel(window_initial, size_final, list_strides)

Return size of kernel according to : window_initial : size of window size_final : size final list_strides : list of strides

return(list_kernel,list_strides)

uqmodels.modelization.DL_estimator.utils.identity(*args)

uqmodels.modelization.DL_estimator.utils.set_global_determinism(seed=None)

uqmodels.modelization.DL_estimator.utils.set_seeds(seed=None)

uqmodels.modelization.DL_estimator.utils.size_post_conv(w, l_k, l_st)

provide size post conv (with padding=valid) w : size of window l_k : list kernel l_s : list_stride

uqmodels.modelization.DL_estimator.utils.sum_part_prod(array)

compute sum_part_prod array = [k1,…,kn] return (k1+k1k2+k1k2k3+..+k1..Kn)

Module contents