stable_learning_control.algos.pytorch.common.helpers

Contains several Pytorch helper functions.

Functions

retrieve_device([device_type])	Retrieves the available computational device given a device type.
mlp(sizes, activation[, output_activation])	Create a multi-layered perceptron using pytorch.
count_vars(module)	Returns the total number of parameters of a pytorch module.
compare_models(model_1, model_2)	Compares two models to see if the weights are equal.
rescale(data, min_bound, max_bound)	Rescale normalized data (i.e. between -1 and 1) to a desired range.
np_to_torch(input_object[, dtype, device])	Converts all numpy arrays in a python object to Torch Tensors.

Module Contents

stable_learning_control.algos.pytorch.common.helpers.retrieve_device(device_type='cpu')

Retrieves the available computational device given a device type.

Parameters:

device_type (str) – The device type (options: cpu, gpu, gpu:0, gpu:1, etc.). Defaults to cpu.

Returns:

The Pytorch device object.

Return type:

torch.device

stable_learning_control.algos.pytorch.common.helpers.mlp(sizes, activation, output_activation=nn.Identity)

Create a multi-layered perceptron using pytorch.

Parameters:

• sizes (list) – The size of each of the layers.

• activation (union[torch.nn.modules.activation, str]) – The activation function used for the hidden layers.

• output_activation (union[torch.nn.modules.activation, str], optional) – The activation function used for the output layers. Defaults to torch.nn.Identity.

Returns:

The multi-layered perceptron.

Return type:

torch.nn.Sequential

stable_learning_control.algos.pytorch.common.helpers.count_vars(module)

Returns the total number of parameters of a pytorch module.

Parameters:

module (torch.nn.Module) – The module.

Returns:

The total number of parameters inside the module.

Return type:

numpy.int64

stable_learning_control.algos.pytorch.common.helpers.compare_models(model_1, model_2)

Compares two models to see if the weights are equal.

Parameters:

• model_1 (torch.nn.Module) – The first Pytorch model.

• model_2 (torch.nn.Module) – The second Pytorch model.

Raises:

Exception – Raises Key error if the graph of the two models is different.

Returns:

Bool specifying whether the weights of two models are equal.

Return type:

bool

stable_learning_control.algos.pytorch.common.helpers.rescale(data, min_bound, max_bound)

Rescale normalized data (i.e. between -1 and 1) to a desired range.

Parameters:

• data (Union[torch.Tensor, numpy.ndarray, list]) – Normalized input data.

• min_bound (Union[numpy.ndarray, list]) – Array containing the minimum value of the desired range.

• max_bound (Union[numpy.ndarray, list]) – Array containing the maximum value of the desired range.

Returns:

Array which has it values scaled between

the min and max boundaries.

Return type:

Union[Torch.Tensor, numpy.ndarray]

stable_learning_control.algos.pytorch.common.helpers.np_to_torch(input_object, dtype=None, device=None)

Converts all numpy arrays in a python object to Torch Tensors.

Parameters:

• input_item (obj) – The python object.

• dtype (type, optional) – The type you want to use for storing the data in the tensor. Defaults to None (i.e. torch default will be used).

• device (str, optional) – The computational device on which the tensors should be stored. (options: cpu, gpu, gpu:0, gpu:1, etc.). Defaults to None (i.e. torch default device will be used).

Returns:

The output python object in which numpy arrays have been converted to

torch tensors.

Return type:

object