mlcolvar.cvs.BaseCV

class mlcolvar.cvs.BaseCV(in_features, out_features, preprocessing: Module = None, postprocessing: Module = None, *args, **kwargs)

Bases: object

Base collective variable class.

To inherit from this class, the class must define a BLOCKS class attribute.

__init__(in_features, out_features, preprocessing: Module = None, postprocessing: Module = None, *args, **kwargs)

Base CV class options.

Parameters:

• in_features (int) – Number of inputs of the CV model

• out_features (int) – Number of outputs of the CV model, should be the number of CVs

• preprocessing (torch.nn.Module, optional) – Preprocessing module, default None

• postprocessing (torch.nn.Module, optional) – Postprocessing module, default None

Methods

__init__(in_features, out_features[, ...])	Base CV class options.
configure_optimizers()	Initialize the optimizer based on self._optimizer_name and self.optimizer_kwargs.
forward(x)	Evaluation of the CV
forward_cv(x)	Execute sequentially all the blocks in self.BLOCKS unless they are not initialized.
initialize_blocks()	Initialize the blocks as attributes of the CV class.
initialize_transforms(datamodule)
parse_options([options])	Sanitize options and create defaults ({}) if not in options.
setup([stage])
test_step(test_batch, batch_idx)	Equal to training step if not overridden.
validation_step(val_batch, batch_idx)	Equal to training step if not overridden.

configure_optimizers()

Initialize the optimizer based on self._optimizer_name and self.optimizer_kwargs. It also adds the learning rate scheduler if self.lr_scheduler_kwargs is not empty. The scheduler is given as a dictionary with the key ‘scheduler’ containing the scheduler class and the rest of the keys are config options for the scheduler.

Returns:

• torch.optim – Torch optimizer

• dict, optional – Learning rate scheduler configuration (if any)

forward(x: Tensor) → Tensor

Evaluation of the CV

• Apply preprocessing if any

• Execute sequentially all the blocks in self.BLOCKS unless they are not initialized

• Apply postprocessing if any

Parameters:

x (torch.Tensor) – Input of the forward operation of the model

Returns:

Output of the forward operation of the model

Return type:

torch.Tensor

forward_cv(x: Tensor) → Tensor

Execute sequentially all the blocks in self.BLOCKS unless they are not initialized.

No pre/post processing will be executed here. This is supposed to be called during training/validation and to be overloaded if necessary.

Parameters:

x (torch.Tensor) – Input of the forward operation of the model

Returns:

Output of the forward operation of the model

Return type:

torch.Tensor

initialize_blocks()

Initialize the blocks as attributes of the CV class.

property n_cvs

Number of CVs.

property optimizer_name: str

Optimizer name. Options can be set using optimizer_kwargs. Actual optimizer will be return during training from configure_optimizer function.

parse_options(options: dict = None)

Sanitize options and create defaults ({}) if not in options. Furthermore, it sets the optimizer kwargs, if given.

Parameters:

options (dict[str, Any], optional) – Options for the building blocks of the model, by default None.

test_step(test_batch, batch_idx)

Equal to training step if not overridden. Different behaviors for train/valid step can be enforced in training_step() based on the self.training variable.

validation_step(val_batch, batch_idx)

Equal to training step if not overridden. Different behaviors for train/valid step can be enforced in training_step() based on the self.training variable.

Attributes

example_input_array
n_cvs	Number of CVs.
optimizer_name	Optimizer name.