Closed chadcwilliams closed 1 year ago
Turns out this was not specific to one electrode - the AE was broken. This happened because we weren't saving num_layers in the configuration and the AE system_inputs default was 2, but the AE class defaults to 3 (and when loading the AE for GAN, it does not pull defaults from system_inputs. Fixed in main now.
When using the DoubleAE, the training works fine if you have one electrode, but it fails to load later saying it is missing the second layer parameters in the configuration dict.