Overview
Baseline is a simple last-value-repeat forecasting model that uses the last known encoder target value as the prediction for all future time steps.
Description
Baseline extends BaseModel and implements the simplest possible forecasting strategy: it takes the last observed value from the encoder target and repeats it across the entire prediction horizon. It supports both single-target and multi-target scenarios. The model requires no trainable parameters and serves as a reference point for evaluating the performance of more sophisticated models. It also provides to_prediction and to_quantiles convenience methods.
Usage
Use Baseline as a naive benchmark model to establish a performance floor when developing and evaluating forecasting models. It is useful for sanity-checking data pipelines and for comparing against learned models to quantify the value of model complexity.
Code Reference
Source Location
Signature
class Baseline(BaseModel):
def forward(self, x: dict[str, torch.Tensor]) -> dict[str, torch.Tensor]:
def forward_one_target(
self,
encoder_lengths: torch.Tensor,
decoder_lengths: torch.Tensor,
encoder_target: torch.Tensor,
):
def to_prediction(self, out: dict[str, Any], use_metric: bool = True, **kwargs):
def to_quantiles(self, out: dict[str, Any], use_metric: bool = True, **kwargs):
Import
from pytorch_forecasting.models.baseline import Baseline
I/O Contract
Inputs
forward
| Name |
Type |
Required |
Description
|
| x |
dict[str, torch.Tensor] |
Yes |
Network input dictionary containing "encoder_target" (torch.Tensor or list of torch.Tensor), "encoder_lengths" (torch.Tensor), and "decoder_lengths" (torch.Tensor).
|
forward_one_target
| Name |
Type |
Required |
Description
|
| encoder_lengths |
torch.Tensor |
Yes |
Tensor of encoder sequence lengths per sample in the batch.
|
| decoder_lengths |
torch.Tensor |
Yes |
Tensor of decoder (prediction) sequence lengths per sample in the batch.
|
| encoder_target |
torch.Tensor |
Yes |
Encoder target values tensor of shape (batch_size, max_encoder_length).
|
to_prediction
| Name |
Type |
Required |
Description
|
| out |
dict[str, Any] |
Yes |
Network output dictionary (from forward).
|
| use_metric |
bool |
No |
Whether to use metric for conversion. Defaults to True.
|
to_quantiles
| Name |
Type |
Required |
Description
|
| out |
dict[str, Any] |
Yes |
Network output dictionary (from forward).
|
| use_metric |
bool |
No |
Whether to use metric for conversion. Defaults to True.
|
Outputs
forward
| Name |
Type |
Description
|
| output |
dict[str, torch.Tensor] |
Network output dictionary with "prediction" key containing the repeated last-value tensor of shape (batch_size, max_prediction_length) or a list thereof for multi-target.
|
forward_one_target
| Name |
Type |
Description
|
| prediction |
torch.Tensor |
Prediction tensor of shape (batch_size, max_prediction_length) where each row is the last encoder value repeated.
|
to_prediction
| Name |
Type |
Description
|
| prediction |
torch.Tensor |
The raw prediction tensor from the output.
|
to_quantiles
| Name |
Type |
Description
|
| quantiles |
torch.Tensor |
The prediction tensor with an extra trailing dimension: (batch_size, max_prediction_length, 1).
|
Usage Examples
import torch
from pytorch_forecasting.models.baseline import Baseline
from pytorch_forecasting.metrics import MAE
# Generate predictions using a dataloader
model = Baseline()
predictions = model.predict(dataloader)
# Calculate baseline MAE performance
metric = MAE()
for x, y in dataloader:
metric.update(model(x), y)
result = metric.compute()
Related Pages
Page Connections
Double-click a node to navigate. Hold to expand connections.
