Train multiple connected neural networks with a single optimizer

Question

How can I jointly optimize the parameters of a model comprising two distinct neural networks with a single optimizer? What I've tried is the following, after having initialized an optimizer:

optim_global = optim.Adam(zip(model1.parameters(), model2.parameters()))

but I get this error

TypeError: optimizer can only optimize Tensors, but one of the params is tuple

Answer 1

These are generator you can control either with the unpacking operator *:

>>> optim.Adam([*model1.parameters(), *model2.parameters()])

Or using itertools.chain

>>> optim.Adam(chain(model1.parameters(), model2.parameters()))