torchsig.utils.data_loading.metadata_padding_collate_fn

torchsig.utils.data_loading.metadata_padding_collate_fn(batch)[source]

Collate a batch of (data, metadata_list) pairs, padding metadata to equal lengths.

Metadata for each sample is a list of dicts. This function:

  1. Finds the maximum metadata-list length in the batch.

  2. Pads shorter metadata lists with default values.

  3. Stacks data tensors and metadata fields into batched tensors.

Parameters:

batch – A list where each element is a tuple of: - x: any object convertible to a NumPy array (e.g., tensor, array). - y: a list of metadata dicts, where each dict shares the same set of keys.

Returns:

  • data_tensor: stacked torch.Tensor of all x values, shape (batch_size, …).

  • metadata_tensors: dict mapping each metadata key to a Tensor of shape (batch_size, max_sequence_length).

Return type:

A tuple containing

Raises:

ValueError – if any element in batch is not a tuple of length 2.