I’m trying to understand why PyTorch’s nn.Linear(in_features, out_features) layer has its weight matrix with the shape (out_features, in_features) instead of (in_features, out_features).

I am relatively new to Pytorch and was wondering if I could get any recommendations on tracking down a gradient flow issue. The initial back-propagation shows flow throughout the CG:

I have a beginner question, which I can’t seem to find a definitive answer to anywhere.

I tried to reduce my problem to the following code:

I’m using this approximation example for a sin wave as basis to learning approximation. I’m new to PyTorch… Why do different BATCH_SIZE values significantly change results?