DeepStateNet

class DeepStateNet(ssm: etna.models.nn.deepstate.state_space_model.CompositeSSM, input_size: int, num_layers: int, n_samples: int, lr: float, optimizer_params: Optional[dict])

Bases: etna.models.base.DeepBaseNet

DeepState network.

Create instance of DeepStateNet.

Parameters

• ssm (etna.models.nn.deepstate.state_space_model.CompositeSSM) – State Space Model of the system.

• input_size (int) – Size of the input feature space: features for RNN part.

• num_layers (int) – Number of layers in RNN.

• n_samples (int) – Number of samples to use in predictions generation.

• lr (float) – Learning rate.

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

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.deepstate.deepstate.DeepStateBatch, *args, **kwargs)

Forward pass.

Parameters

x (etna.models.nn.deepstate.deepstate.DeepStateBatch) – 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.deepstate.deepstate.DeepStateBatch, *args, **kwargs)

Step for loss computation for training or validation.

Parameters

batch (etna.models.nn.deepstate.deepstate.DeepStateBatch) – batch of data

Returns

loss, true_target, prediction_target