def train_overfit_classifier(num_batches, batch_size):
print("Training mnist classifier...")
next_batch_iter = mnist_utils.get_two_class_iterator(
'train',
num_datapoints=num_batches * batch_size,
batch_size=batch_size,
label_scheme='one_hot',
cycle=True)
mnist_utils.train_mnist(model_dir,
lambda: next(next_batch_iter),
num_batches,
"vanilla",
save_every=(num_batches - 1),
print_every=10)
return mnist_utils.np_two_class_mnist_model(model_dir)
def train_overfit_classifier(num_batches, batch_size):
mnist = mnist_utils.mnist_dataset(one_hot=False)
print("Training mnist classifier...")
next_batch_iter = mnist_utils.two_class_iter(
mnist.train.images, mnist.train.labels,
num_datapoints=num_batches * batch_size,
batch_size=batch_size,
label_scheme='one_hot',
cycle=True)
mnist_utils.train_mnist(
model_dir,
lambda: next(next_batch_iter),
num_batches, "vanilla",
save_every=(num_batches - 1),
print_every=10)
return mnist_utils.np_two_class_mnist_model(model_dir)
def main(_):
assert FLAGS.train_mode in ['vanilla', 'adversarial']
mnist = mnist_utils.mnist_dataset(one_hot=True)
next_batch_fn = lambda: mnist.train.next_batch(FLAGS.batch_size)
mnist_utils.train_mnist(FLAGS.model_dir, next_batch_fn,
FLAGS.total_batches, FLAGS.train_mode)