PatchTSNet

class PatchTSNet(encoder_length: int, patch_len: int, stride: int, num_layers: int, hidden_size: int, feedforward_size: int, nhead: int, lr: float, loss: torch.nn.modules.module.Module, optimizer_params: Optional[dict])

Bases: etna.models.base.DeepBaseNet

PatchTS based Lightning module.

Init PatchTS.

Parameters

• encoder_length (int) – encoder length

• patch_len (int) – size of patch

• stride (int) – step of patch

• num_layers (int) – number of layers

• hidden_size (int) – size of the hidden state

• feedforward_size (int) – size of feedforward layers in transformer

• nhead (int) – number of transformer heads

• lr (float) – learning rate

• loss (torch.nn.Module) – loss function

• optimizer_params (Optional[dict]) – parameters for optimizer for Adam optimizer (api reference torch.optim.Adam)

Return type

None

Methods

configure_optimizers()	Optimizer configuration.
forward(x, *args, **kwargs)	Forward pass.
make_samples(df, encoder_length, decoder_length)	Make samples from segment DataFrame.
step(batch, *args, **kwargs)	Step for loss computation for training or validation.

Attributes

configure_optimizers() → torch.optim.optimizer.Optimizer

Optimizer configuration.

Return type

torch.optim.optimizer.Optimizer

forward(x: etna.models.nn.patchts.PatchTSBatch, *args, **kwargs)

Forward pass.

Parameters

x (etna.models.nn.patchts.PatchTSBatch) – batch of data

Returns

forecast with shape (batch_size, decoder_length, 1)

make_samples(df: pandas.core.frame.DataFrame, encoder_length: int, decoder_length: int) → Iterator[dict]

Make samples from segment DataFrame.

Parameters

• df (pandas.core.frame.DataFrame) –

• encoder_length (int) –

• decoder_length (int) –

Return type

Iterator[dict]

step(batch: etna.models.nn.patchts.PatchTSBatch, *args, **kwargs)

Step for loss computation for training or validation.

Parameters

batch (etna.models.nn.patchts.PatchTSBatch) – batch of data

Returns

loss, true_target, prediction_target