I have a list of tensors and I have say 500 classes, I want to calculate the mean vector for each class in a differentiable way.
here is my code:
y is the labels
H =torch.randn (800,800)
C = 500 mean_vectors_list =  mean_vectors_list.append([torch.mean(H[y == cl], dim=0, dtype=torch.float32) for cl in range(C)])
The problem is some times H tensor doesn’t have some classes as it’s a randomly selected batch
that gives me Nan.
How to use .scatter_add_ to produce the desired output with cuda