16. Training Two Networks

eval

Note that the Generator should be set to eval mode for sample generation. It doesn't make too big a difference in this case, but this is to account for the different behavior that a dropout layer has during training vs during model evaluation.

So, in the workspace and Github repository code, we've added the correct evaluation code for generating samples, writing the line G.eval() before we generate samples. We strive to always keep the code that you'll be working with correct and up-to-date!