The chytorch's discuss from chython

chytorch's Issues

Collate import incompatible with torch version <1.13

One import makes chytorch incompatible with PyTorch with a version smaller than 1.13.

The import is located in utils/data/molecule/contrastive.py (line 25):

from torch.utils.data._utils.collate import default_collate_fn_map

In torch 1.13 dict default_collate_fn_map can be found here: https://github.com/pytorch/pytorch/blob/v1.13.0/torch/utils/data/_utils/collate.py#L191 , however in older versions it is not present.

It seems that it is enough to copy-paste the code from PyTorch 1.13 that is responsible for creating default_collate_fn_map:

default_collate_fn_map: Dict[Union[Type, Tuple[Type, ...]], Callable] = {torch.Tensor: collate_tensor_fn}
with contextlib.suppress(ImportError):
    import numpy as np
    # For both ndarray and memmap (subclass of ndarray)
    default_collate_fn_map[np.ndarray] = collate_numpy_array_fn
    # See scalars hierarchy: https://numpy.org/doc/stable/reference/arrays.scalars.html
    # Skip string scalars
    default_collate_fn_map[(np.bool_, np.number, np.object_)] = collate_numpy_scalar_fn
default_collate_fn_map[float] = collate_float_fn
default_collate_fn_map[int] = collate_int_fn
default_collate_fn_map[string_classes] = collate_str_fn

Although I do not see any usage of the default_collate_fn_map in the repo... Is it relevant to have it?

Recommend Projects