search

tolstikhin / adagan

AdaGAN: greedy iterative procedure to train mixtures of GANs
BSD 3-Clause "New" or "Revised" License
76 stars 18 forks source link

Why replace=False? #4

Open a554b554 opened 5 years ago

a554b554 commented 5 years ago

In gan.py line 401, data_ids = np.random.choice(train_size, opts['batch_size'], replace=False, p=self._data_weights) Why use replace=False here?

tolstikhin commented 5 years ago

Well, I guess sampling minibatches with and without replacements are both common approaches. Here we used without replacements.