Minor edits to enable compatibility with recent torch versions

[colleenjg]

Fixes #25, enabling torch compatibility (checked up to 18.1).

dpc/main.py, lines 215 and 267:

• target_flattened placed on cuda when created, to avoid an error when criterion() is called.
• target_flattened converted to int before argmax() to avoid an error with if target_flattened is a boolean tensor.

utils/utils.py, line 53:

• In correct_k = correct[:k].view(-1).float().sum(0), a contiguous() call is added to avoid an error when view(-1) is called.

[TengdaHan]

Thanks!