I’m trying to train a classifier which starts with n outputs (n classes) from a linear layer,.

After some processes, I want to retrain it with the same weights on each layer, but the output layer needs to expand to n+1 outputs, using the same trained weights.

Is there a way to expand the weights?, Can I for example, copypaste the output weights from a save of the original model to a new array of the new size and use them as the new weights?

I’m fairly new on PyTorch and I haven’t found if there’s something similar to what i’m trying to do.

Thanks!