*TFT The Temporal Fusion Transformer architecture (TFT) is an Sequence-to-Sequence model that combines static, historic and future available data to predict an univariate target. The method combines gating layers, an LSTM recurrent encoder, with and interpretable multi-head attention layer and a multi-step forecasting strategy decoder. Parameters:
h
: int, Forecast horizon. input_size
: int,
autorregresive inputs size, y=[1,2,3,4] input_size=2 ->
y_[t-2:t]=[1,2].tgt_size
: int=1, target size.stat_exog_list
: str list, static continuous columns.hist_exog_list
: str list, historic continuous columns.futr_exog_list
: str list, future continuous columns.hidden_size
: int, units of embeddings and encoders.n_head
:
int=4, number of attention heads in temporal fusion decoder.attn_dropout
: float (0, 1), dropout of fusion decoder’s attention
layer.grn_activation
: str, activation for the GRN module from
[‘ReLU’, ‘Softplus’, ‘Tanh’, ‘SELU’, ‘LeakyReLU’, ‘Sigmoid’, ‘ELU’,
‘GLU’].n_rnn_layers
: int=1, number of RNN layers.rnn_type
: str=“lstm”, recurrent neural network (RNN) layer type from
[“lstm”,“gru”].one_rnn_initial_state
:str=False, Initialize all
rnn layers with the same initial states computed from static
covariates.dropout
: float (0, 1), dropout of inputs VSNs.loss
: PyTorch module, instantiated train loss class from losses
collection.valid_loss
: PyTorch module=loss
, instantiated valid loss class from
losses
collection.max_steps
: int=1000, maximum number of training steps.learning_rate
: float=1e-3, Learning rate between (0, 1).num_lr_decays
: int=-1, Number of learning rate decays, evenly
distributed across max_steps.early_stop_patience_steps
: int=-1,
Number of validation iterations before early stopping.val_check_steps
: int=100, Number of training steps between every
validation loss check.batch_size
: int, number of different series
in each batch.valid_batch_size
: int=None, number of different
series in each validation and test batch.windows_batch_size
:
int=None, windows sampled from rolled data, default uses all.inference_windows_batch_size
: int=-1, number of windows to sample in
each inference batch, -1 uses all.start_padding_enabled
:
bool=False, if True, the model will pad the time series with zeros at
the beginning, by input size.step_size
: int=1, step size between
each window of temporal data.scaler_type
: str=‘robust’, type of
scaler for temporal inputs normalization see temporal
scalers.random_seed
: int, random seed initialization for replicability.drop_last_loader
: bool=False, if True TimeSeriesDataLoader
drops
last non-full batch.alias
: str, optional, Custom name of the
model.optimizer
: Subclass of ‘torch.optim.Optimizer’, optional,
user specified optimizer instead of the default choice (Adam).optimizer_kwargs
: dict, optional, list of parameters used by the user
specified optimizer
.lr_scheduler
: Subclass of
‘torch.optim.lr_scheduler.LRScheduler’, optional, user specified
lr_scheduler instead of the default choice (StepLR).lr_scheduler_kwargs
: dict, optional, list of parameters used by the
user specified lr_scheduler
.dataloader_kwargs
: dict, optional,
list of parameters passed into the PyTorch Lightning dataloader by the
TimeSeriesDataLoader
. **trainer_kwargs
: int, keyword trainer
arguments inherited from PyTorch Lighning’s
trainer.*Fit. The
fit
method, optimizes the neural network’s weights using the
initialization parameters (learning_rate
, windows_batch_size
, …) and
the loss
function as defined during the initialization. Within fit
we use a PyTorch Lightning Trainer
that inherits the initialization’s
self.trainer_kwargs
, to customize its inputs, see PL’s trainer
arguments.
The method is designed to be compatible with SKLearn-like classes and in
particular to be compatible with the StatsForecast library.
By default the model
is not saving training checkpoints to protect
disk memory, to get them change enable_checkpointing=True
in
__init__
.
Parameters:dataset
: NeuralForecast’s
TimeSeriesDataset
,
see
documentation.val_size
: int, validation size for temporal cross-validation.random_seed
: int=None, random_seed for pytorch initializer and numpy
generators, overwrites model.__init__’s.test_size
: int, test
size for temporal cross-validation.*Predict. Neural network prediction with PL’s
Trainer
execution of
predict_step
.
Parameters:dataset
: NeuralForecast’s
TimeSeriesDataset
,
see
documentation.test_size
: int=None, test size for temporal cross-validation.step_size
: int=1, Step size between each window.random_seed
:
int=None, random_seed for pytorch initializer and numpy generators,
overwrites model.__init__’s.quantiles
: list of floats,
optional (default=None), target quantiles to predict. **data_module_kwargs
: PL’s TimeSeriesDataModule args, see
documentation.*
*Compute the feature importances for historical, future, and static features. Returns: dict: A dictionary containing the feature importances for each feature type. The keys are ‘hist_vsn’, ‘future_vsn’, and ‘static_vsn’, and the values are pandas DataFrames with the corresponding feature importances.*
*Batch average attention weights Returns: np.ndarray: A 1D array containing the attention weights for each time step.*
*Batch average attention weights Returns: np.ndarray: A 1D array containing the attention weights for each time step.*
*Compute the correlation between the past and future feature importances and the mean attention weights. Returns: pd.DataFrame: A DataFrame containing the correlation coefficients between the past feature importances and the mean attention weights.*