Chap4 (CNN_working_details.ipynb) -- Sumprod (Device: CPU) X & cnn_w (Device: GPU ("cuda" or "mps"))

[Camaltra]

In the code

sumprod = torch.zeros((h_im - h_conv + 1, w_im - w_conv + 1))

for i in range(h_im - h_conv + 1):
    for j in range(w_im - w_conv + 1):
        img_subset = X_train[0, 0, i:(i+3), j:(j+3)]
        model_filter = cnn_w.reshape(3,3)
        val = torch.sum(img_subset*model_filter) + cnn_b
        sumprod[i,j] = val

Sumprod is on the CPU device, while X_train and model_filter that come from the cnn_w form the model are both init on the GPU

So when come the assignment, there will have an issues, and the sumprod will look like this

tensor([[val, 0],
              [0, 0]])

To fix it, just specify the device to sumprod

sumprod = torch.zeros((h_im - h_conv + 1, w_im - w_conv + 1)).to(device)