GPU Support for Mac Users

[ramosv]

Mac now support Accelerated PyTorch training on Mac via their M-series chips see: https://developer.apple.com/metal/pytorch/

Would be nice for mac users to leverage their GPU for accelerated training. I will be working on adding this before v1.3.0 release

[ramosv]

Example of what would need to be added: The middle elif is all we need to add to the existing block

if torch.cuda.is_available():
    device = torch.device("cuda")
elif torch.backends.mps.is_available():
    device = torch.device("mps")
else:
    device = torch.device("cpu")

[ramosv]

I reviewed the codebase to identify the specific locations where we need to implement this update

Several files rely on a hardcoded pattern similar to this:
torch.device("cuda" if self.gpu and torch.cuda.is_available() else "cpu")

To improve readability and properly support MPS, we should replace these instances with a standardized logic block. Ideally, we can abstract this into a single utility function at the root of each file or class.

if torch.cuda.is_available():
    device = torch.device("cuda")
elif torch.backends.mps.is_available():
    device = torch.device("mps")
else:
    device = torch.device("cpu")

Files and Locations to Update

Here are the specific files and line numbers that will need this update:

Downstream Tasks:

• bioneuralnet/downstream_task/dpmon.py: Update the setup_device(gpu, cuda) function at line 323.
• bioneuralnet/downstream_task/subject_representation.py: Line 398 currently uses resources = {"cpu": 1, "gpu": 1} if torch.cuda.is_available() else {"cpu": 1, "gpu": 0}. It would be best to replace this entirely and refactor it to use the standardized setup function.

Network Generation & Tools:

• bioneuralnet/network/generate.py: This affects all four network types at lines 30, 115, 217, and 288.
• bioneuralnet/network/tools.py: Update the __init__ function for the NetworkAnalyzer class at line 42.

PySmCCNet:

• bioneuralnet/network/pysmccnet/analysis.py: Update lines 45 and 85.
• bioneuralnet/network/pysmccnet/pipeline.py: Update line 62.

Embeddings & Utils:

• bioneuralnet/network_embedding/gnn_embedding.py: Update lines 124 and 145.
• bioneuralnet/utils/reproducibility.py: Ensure the device logic here also accommodates the MPS fallback.