Implementation:Sktime Pytorch forecasting TiDE

Knowledge Sources	Sktime_Pytorch_forecasting
Domains	Time_Series, Forecasting, Deep_Learning
Last Updated	2026-02-08 08:00 GMT

Overview

TiDE (Time-series Dense Encoder) is an MLP-based encoder-decoder model designed for long-term time series forecasting without attention mechanisms.

Description

TiDEModel extends BaseModelWithCovariates and implements the TiDE architecture, which uses dense residual blocks for encoding and decoding instead of attention-based transformers. The model encodes the flattened historical targets along with projected encoder and decoder covariates through stacked residual blocks, then decodes through additional residual blocks and a temporal decoder. A linear skip connection from lookback targets to future predictions adds a direct autoregressive path. The model supports future covariates, static covariates, and multi-target forecasting, while achieving competitive performance with reduced computational cost compared to transformer-based approaches.

Usage

Use TiDE when you need efficient long-term time series forecasting, particularly when computational resources are limited. It provides transformer-competitive accuracy with MLP-based simplicity. The preferred instantiation method is via from_dataset, which automatically configures input/output dimensions from a TimeSeriesDataSet.

Code Reference

Source Location

• Repository: Sktime_Pytorch_forecasting
• File: pytorch_forecasting/models/tide/_tide.py
• Lines: 1-338

Signature

class TiDEModel(BaseModelWithCovariates):
    def __init__(
        self,
        output_chunk_length: int,
        input_chunk_length: int,
        num_encoder_layers: int = 2,
        num_decoder_layers: int = 2,
        decoder_output_dim: int = 16,
        hidden_size: int = 128,
        temporal_width_future: int = 4,
        temporal_hidden_size_future: int = 32,
        temporal_decoder_hidden: int = 32,
        use_layer_norm: bool = False,
        dropout: float = 0.1,
        output_size: int | list[int] = 1,
        static_categoricals: list[str] | None = None,
        static_reals: list[str] | None = None,
        time_varying_categoricals_encoder: list[str] | None = None,
        time_varying_categoricals_decoder: list[str] | None = None,
        categorical_groups: dict[str, list[str]] | None = None,
        time_varying_reals_encoder: list[str] | None = None,
        time_varying_reals_decoder: list[str] | None = None,
        embedding_sizes: dict[str, tuple[int, int]] | None = None,
        embedding_paddings: list[str] | None = None,
        embedding_labels: list[str] | None = None,
        x_reals: list[str] | None = None,
        x_categoricals: list[str] | None = None,
        logging_metrics: nn.ModuleList = None,
        **kwargs,
    ):

from_dataset Signature

@classmethod
def from_dataset(cls, dataset: TimeSeriesDataSet, **kwargs):

Import

from pytorch_forecasting.models.tide import TiDEModel

I/O Contract

Inputs

Name	Type	Required	Description
output_chunk_length	int	Yes	Number of time steps the model predicts simultaneously (horizon).
input_chunk_length	int	Yes	Number of past time steps used as input (lookback).
num_encoder_layers	int	No	Number of residual blocks in the encoder. Defaults to 2.
num_decoder_layers	int	No	Number of residual blocks in the decoder. Defaults to 2.
decoder_output_dim	int	No	Dimensionality of the decoder output. Defaults to 16.
hidden_size	int	No	Size of hidden layers in encoder/decoder. Defaults to 128.
temporal_width_future	int	No	Output width for future covariate projections (0 = bypass). Defaults to 4.
temporal_hidden_size_future	int	No	Hidden width for future covariate projections. Defaults to 32.
temporal_decoder_hidden	int	No	Width of temporal decoder layers. Defaults to 32.
use_layer_norm	bool	No	Whether to apply layer normalization. Defaults to False.
dropout	float	No	Dropout probability. Defaults to 0.1.
output_size	list[int]	No	Number of outputs per target. Defaults to 1.
logging_metrics	nn.ModuleList	No	Metrics for logging. Defaults to [SMAPE, MAE, RMSE, MAPE, MASE].
static_categoricals	None	No	Names of static categorical variables.
static_reals	None	No	Names of static continuous variables.
embedding_sizes	None	No	Categorical variable embedding sizes.

Outputs

Name	Type	Description
prediction	torch.Tensor	Forecast output transformed to target space, shape (batch_size, output_chunk_length, output_dim) or list of tensors for multi-target.

Usage Examples

from pytorch_forecasting import TimeSeriesDataSet
from pytorch_forecasting.models.tide import TiDEModel

# Create model from dataset (preferred)
model = TiDEModel.from_dataset(
    dataset,
    hidden_size=128,
    num_encoder_layers=2,
    num_decoder_layers=2,
    dropout=0.1,
)

# Direct instantiation
model = TiDEModel(
    output_chunk_length=24,
    input_chunk_length=96,
    num_encoder_layers=3,
    num_decoder_layers=3,
    hidden_size=256,
    decoder_output_dim=32,
    temporal_width_future=4,
    temporal_decoder_hidden=64,
    dropout=0.1,
    use_layer_norm=True,
)

Related Pages

• Principle:Sktime_Pytorch_forecasting_Dense_Encoder_Decoder
• Sktime_Pytorch_forecasting_TiDE_V2 - V2 implementation of TiDE on the new data pipeline
• Sktime_Pytorch_forecasting_RecurrentNetwork - Alternative recurrent architecture