graphorge.utilities.data_scalers.TorchMinMaxScaler

class TorchMinMaxScaler(n_features, minimum=None, maximum=None, device_type='cpu')

Bases: object

PyTorch tensor min-max data scaler.

_n_features

Number of features to normalize.

Type:

int

_minimum

Features normalization minimum tensor stored as a torch.Tensor with shape (n_features,).

Type:

torch.Tensor

_maximum

Features normalization maximum tensor stored as a torch.Tensor with shape (n_features,).

Type:

torch.Tensor

_device

Device on which torch.Tensor is allocated.

Type:

torch.device

set_device(self, device_type)

Set device on which torch.Tensor is allocated.

set_minimum(self, minimum)

Set features normalization minimum tensor.

set_maximum(self, maximum)

Set features normalization maximum tensor.

set_minimum_and_maximum(self, minimum, maximum)

Set features normalization minimum and maximum tensors.

fit(self, tensor)

Fit features normalization minimum and maximum tensors.

transform(self, tensor, is_check_data=False)

Normalize features tensor.

inverse_transform(self, tensor, is_check_data=False)

Denormalize features tensor.

_check_minimum(self, minimum)

Check features normalization minimum tensor.

_check_maximum(self, maximum)

Check features normalization maximum tensor.

_check_tensor(self, tensor)

Check features tensor to be transformed.

Constructor.

Parameters:

• n_features (int) – Number of features to scale.

• minimum (torch.Tensor) – Features normalization minimum tensor stored as a torch.Tensor with shape (n_features,).

• maximum (torch.Tensor) – Features normalization maximum tensor stored as a torch.Tensor with shape (n_features,).

• device_type ({'cpu', 'cuda'}, default='cpu') – Type of device on which torch.Tensor is allocated.

List of Public Methods

fit	Fit features normalization minimum and maximum tensors.
inverse_transform	Denormalize features tensor.
set_device	Set device on which torch.Tensor is allocated.
set_maximum	Set features normalization maximum tensor.
set_minimum	Set features normalization minimum tensor.
set_minimum_and_maximum	Set features normalization minimum and maximum tensors.
transform	Normalize features tensor.

Methods

__init__(n_features, minimum=None, maximum=None, device_type='cpu')

Constructor.

Parameters:

• n_features (int) – Number of features to scale.

• minimum (torch.Tensor) – Features normalization minimum tensor stored as a torch.Tensor with shape (n_features,).

• maximum (torch.Tensor) – Features normalization maximum tensor stored as a torch.Tensor with shape (n_features,).

• device_type ({'cpu', 'cuda'}, default='cpu') – Type of device on which torch.Tensor is allocated.

_check_maximum(maximum)

Check features normalization maximum tensor.

Parameters:

maximum (torch.Tensor) – Features normalization maximum tensor stored as a torch.Tensor with shape (n_features,).

Returns:

maximum – Features normalization maximum tensor stored as a torch.Tensor with shape (n_features,).

Return type:

torch.Tensor

_check_minimum(minimum)

Check features normalization minimum tensor.

Parameters:

minimum (torch.Tensor) – Features normalization minimum tensor stored as a torch.Tensor with shape (n_features,).

Returns:

minimum – Features normalization minimum tensor stored as a torch.Tensor with shape (n_features,).

Return type:

torch.Tensor

_check_tensor(tensor)

Check features tensor to be transformed.

Parameters:

tensor (torch.Tensor) – Features PyTorch tensor stored as torch.Tensor with shape (n_samples, n_features) or as torch.Tensor with shape (sequence_length, n_samples, n_features).

fit(tensor)

Fit features normalization minimum and maximum tensors.

If sequential data is provided, then all sequence times are concatenated and the normalization procedures take into account the whole sequence length data for each feature.

Parameters:

tensor (torch.Tensor) – Features PyTorch tensor stored as torch.Tensor with shape (n_samples, n_features) or as torch.Tensor with shape (sequence_length, n_samples, n_features).

inverse_transform(tensor, is_check_data=False)

Denormalize features tensor.

If sequential data is provided, then all sequence times are concatenated and the normalization procedures take into account the whole sequence length data for each feature.

Parameters:

• tensor (torch.Tensor) – Normalized features PyTorch tensor stored as torch.Tensor with shape (n_samples, n_features) or as torch.Tensor with shape (sequence_length, n_samples, n_features).

• is_check_data (bool, default=False) – If True, then check transformed tensor data.

Returns:

transformed_tensor – Features PyTorch tensor stored as torch.Tensor with shape (n_samples, n_features) or as torch.Tensor with shape (sequence_length, n_samples, n_features).

Return type:

torch.Tensor

set_device(device_type)

Set device on which torch.Tensor is allocated.

Parameters:

• device_type ({'cpu', 'cuda'}) – Type of device on which torch.Tensor is allocated.

• device (torch.device) – Device on which torch.Tensor is allocated.

set_maximum(maximum)

Set features normalization maximum tensor.

Parameters:

maximum (torch.Tensor) – Features normalization maximum tensor stored as a torch.Tensor with shape (n_features,).

set_minimum(minimum)

Set features normalization minimum tensor.

Parameters:

minimum (torch.Tensor) – Features normalization minimum tensor stored as a torch.Tensor with shape (n_features,).

set_minimum_and_maximum(minimum, maximum)

Set features normalization minimum and maximum tensors.

Parameters:

• minimum (torch.Tensor) – Features normalization minimum tensor stored as a torch.Tensor with shape (n_features,).

• maximum (torch.Tensor) – Features normalization maximum tensor stored as a torch.Tensor with shape (n_features,).

transform(tensor, is_check_data=False)

Normalize features tensor.

If sequential data is provided, then all sequence times are concatenated and the normalization procedures take into account the whole sequence length data for each feature.

Parameters:

• tensor (torch.Tensor) – Features PyTorch tensor stored as torch.Tensor with shape (n_samples, n_features) or as torch.Tensor with shape (sequence_length, n_samples, n_features).

• is_check_data (bool, default=False) – If True, then check transformed tensor data.

Returns:

transformed_tensor – Normalized features PyTorch tensor stored as torch.Tensor with shape (n_samples, n_features) or as torch.Tensor with shape (sequence_length, n_samples, n_features).

Return type:

torch.Tensor