def test_emnist_frechet_distance(self):
distance = eeu.emnist_frechet_distance(
self.real_images, self.fake_images,
ecm.get_trained_emnist_classifier_model())
self.assertAllClose(distance, 568.6883, rtol=0.0001, atol=0.0001)
distance = eeu.emnist_frechet_distance(
self.real_images, self.real_images,
ecm.get_trained_emnist_classifier_model())
self.assertAllClose(distance, 0.0)
def _compute_eval_metrics(generator, discriminator, gen_inputs, real_images,
gan_loss_fns, emnist_classifier):
"""Computes eval metrics for the GAN."""
gen_images = generator(gen_inputs, training=False)
disc_on_real_images = discriminator(real_images, training=False)
disc_on_gen_outputs = discriminator(gen_images, training=False)
real_data_logits = tf.reduce_mean(disc_on_real_images)
gen_data_logits = tf.reduce_mean(disc_on_gen_outputs)
gen_loss = gan_loss_fns.generator_loss(generator, discriminator,
gen_inputs)
disc_loss = gan_loss_fns.discriminator_loss(generator, discriminator,
gen_inputs, real_images)
classifier_score = eeu.emnist_score(gen_images, emnist_classifier)
frechet_classifier_distance = eeu.emnist_frechet_distance(
real_images, gen_images, emnist_classifier)
metrics = collections.OrderedDict([
('real_data_logits', real_data_logits),
('gen_data_logits', gen_data_logits),
('gen trainable norm',
tf.linalg.global_norm(generator.trainable_variables)),
('disc trainable norm',
tf.linalg.global_norm(discriminator.trainable_variables)),
('gen non-trainable norm',
tf.linalg.global_norm(generator.non_trainable_variables)),
('disc non-trainable norm',
tf.linalg.global_norm(discriminator.non_trainable_variables)),
('gen_loss', gen_loss),
('disc_loss', disc_loss),
('classifier_score', classifier_score),
('frechet_classifier_distance', frechet_classifier_distance),
])
return metrics
def _compute_eval_metrics(generator, discriminator, gen_inputs, real_images,
gan_loss_fns, emnist_classifier, server_state):
"""Computes eval metrics for the GAN."""
with tf.device("/cpu:0"):
gen_images = generator(gen_inputs, training=False)
disc_on_real_images = discriminator(real_images, training=False)
disc_on_gen_outputs = discriminator(gen_images, training=False)
real_data_logits = tf.reduce_mean(disc_on_real_images)
gen_data_logits = tf.reduce_mean(disc_on_gen_outputs)
gen_loss = gan_loss_fns.generator_loss(generator, discriminator, gen_inputs)
disc_loss = gan_loss_fns.discriminator_loss(generator, discriminator,
gen_inputs, real_images)
classifier_score = eeu.emnist_score(gen_images, emnist_classifier)
frechet_classifier_distance = eeu.emnist_frechet_distance(
real_images, gen_images, emnist_classifier)
'''
frechet_classifier_distance = eval_util.mnist_frechet_distance(
real_images, gen_images)
'''
metrics = collections.OrderedDict([
('real_data_logits', real_data_logits),
('gen_data_logits', gen_data_logits),
('gen trainable norm',
tf.linalg.global_norm(generator.trainable_variables)),
('disc trainable norm',
tf.linalg.global_norm(discriminator.trainable_variables)),
('gen non-trainable norm',
tf.linalg.global_norm(generator.non_trainable_variables)),
('disc non-trainable norm',
tf.linalg.global_norm(discriminator.non_trainable_variables)),
('gen_loss', gen_loss),
('disc_loss', disc_loss),
('classifier_score', classifier_score),
('frechet_classifier_distance', frechet_classifier_distance),
('gen_opt_state', tf.linalg.global_norm([tf.cast(x, tf.float32) for x in server_state.state_gen_optimizer_weights])),
('disc_opt_state', tf.linalg.global_norm([tf.cast(x, tf.float32) for x in server_state.state_disc_optimizer_weights])),
('gen_diff', tf.linalg.global_norm([tf.cast(x, tf.float32) for x in server_state.generator_diff])),
('disc_diff', tf.linalg.global_norm([tf.cast(x, tf.float32) for x in server_state.discriminator_diff])),
])
return metrics