In the line 177 of "model.py", I think that self.gener_loss should be divided by float(len(scale_weight.keys())). Thus we have the whole average value of the generator. Note that self.gener_acc also gets its mean value by this way. In addition， the discriminator has the same problem in the line 149, self.discr_loss should be divided by float(len(scale_weight.keys())*3).
How do every random batch_size work?
I think initialize_batch_worker(*) in “prepare_dataset.py” put training data into a queue continuously. Then q_art.get() in “model.py” get training data. The training process of a batch size is finished, and then it continue to repeat the previous process. However, I don't know whether this understanding is correct.
I appreciate your work. I'm very grateful for your help.
There are some problems as follows.
In the line 177 of "model.py", I think that self.gener_loss should be divided by float(len(scale_weight.keys())). Thus we have the whole average value of the generator. Note that self.gener_acc also gets its mean value by this way. In addition， the discriminator has the same problem in the line 149, self.discr_loss should be divided by float(len(scale_weight.keys())*3).
How do every random batch_size work? I think initialize_batch_worker(*) in “prepare_dataset.py” put training data into a queue continuously. Then q_art.get() in “model.py” get training data. The training process of a batch size is finished, and then it continue to repeat the previous process. However, I don't know whether this understanding is correct.
I appreciate your work. I'm very grateful for your help.