Updating parameters with two combined network

1

I have an image encoder and a question encoder branch combined together than connecting to 2 fc layers. How should I add the parameter together to update the network parameter?
I have tried some code:

params = model.img_encoder.fc.parameters() + model.qu_encoder.parameters() + \
             model.fc1.parameters() + model.fc2.parameters()

optimizer = optim.Adam(params, lr=1e^-3)

but it will return

TypeError: unsupported operand type(s) for +: 'generator' and 'generator'

any idea or suggestion how to update the parameters of the combined network?

2 
def concat_generators(*args):
      for gen in args:
          yield from gen
  params = concat_generators(model.img_encoder.fc.parameters(), model.qu_encoder.parameters() , model.fc1.parameters(),  model.fc2.parameters())

  optimizer = torch.optim.Adam(params, lr=learning_rate)

This may work.
You got an error because you parameters() are not lists, but generators, and you can’t sum generators.

3

Yes, it work !!! But I think I should watch some tutorial about generator first :joy:

5

Glad it worked! Read stuff about generators, it’s important as they are used everywhere in pytorch :slight_smile:

6

I found anothor solution. The following sample could work, too.

params = list(model.img_encoder.fc.parameters()) + list(model.qu_encoder.parameters()) \
             + list(model.fc1.parameters()) + list(model.fc2.parameters())

optimizer = optim.Adam(params, lr = LEARNING_RATE)