groupby_apply#

pytorch_forecasting.utils._utils.groupby_apply(keys: Tensor, values: Tensor, bins: int = 95, reduction: str = 'mean', return_histogram: bool = False) → Tensor | tuple[Tensor, Tensor][source]#

Groupby apply for torch tensors.

Parameters:

  • keys (torch.Tensor) – Tensor of groups (0 to bins).

  • values (torch.Tensor) – Values to aggregate - same size as keys.

  • bins (int, optional) – Total number of groups. Defaults to 95.

  • reduction (str, optional) – Either “mean” or “sum”. Defaults to “mean”.

  • return_histogram (bool, optional) – If to return histogram on top. Defaults to False.

Returns:

Tensor of size bins with aggregated values and optionally with counts of values.

Return type:

torch.Tensor or tuple of torch.Tensor