def test_metrics(self):
with self.test_session():
graph = mock_gan()
gan = mock_gan()
gan.create()
loss = BoundaryEquilibriumLoss(gan, loss_config)
d_loss, g_loss = loss.create()
metrics = loss.metrics
self.assertTrue(metrics['k'] != None)
def test_create(self):
with self.test_session():
graph = mock_gan()
gan = mock_gan()
gan.create()
loss = LambGanLoss(gan, loss_config)
d_loss, g_loss = loss.create()
d_shape = gan.ops.shape(d_loss)
g_shape = gan.ops.shape(g_loss)
self.assertEqual(d_shape, [])
self.assertEqual(g_shape, [])
def test_batch_norm(self):
with self.test_session():
config['layer_regularizer'] = 'batch_norm'
gan = mock_gan()
generator = ResizableGenerator(config=config, gan=gan, input=gan.latent.sample)
generator.layer_regularizer(tf.constant(1, shape=[1,1,1,1], dtype=tf.float32))
self.assertNotEqual(len(generator.variables()), 0)
def test_train(self):
with self.test_session():
gan = mock_gan()
args = hc.Config({"size": "1", "steps": 1, "method": "train", "save_every": -1})
cli = hg.CLI(gan, args)
cli.train()
self.assertEqual(cli.gan, gan)
def test_run(self):
with self.test_session():
gan = mock_gan()
args = hc.Config({"size": "1"})
cli = hg.CLI(gan, args)
cli.run()
self.assertEqual(cli.gan, gan)
def test_initial_depth(self):
with self.test_session():
gan = mock_gan()
generator = ResizableGenerator(config=config, gan=gan, input=gan.latent.sample)
print(generator.depths())
depths = generator.depths()
self.assertEqual(depths[-1], 2)
def test_cli(self):
with self.test_session():
gan = mock_gan()
args = {
}
cli = hg.CLI(gan, args)
self.assertEqual(cli.gan, gan)
def test_run(self):
with self.test_session():
gan = mock_gan()
args = hc.Config({"size": "1"})
cli = hg.CLI(gan, args)
cli.run()
self.assertEqual(cli.gan, gan)
def test_config(self):
with self.test_session():
gan = mock_gan()
gan.create()
sampler = BatchSampler(gan)
self.assertEqual(sampler._sample()['generator'].shape[-1], 1)
def test_metric(self):
with self.test_session():
gan = mock_gan()
loss = SupervisedLoss(gan, loss_config)
d_loss, g_loss = loss.create()
metrics = loss.metrics
self.assertTrue(metrics['d_class_loss'] != None)
def test_create_component(self):
with self.test_session():
gan = mock_gan()
distribution = gan.create_component(gan.config.latent)
self.assertEqual(
type(distribution),
hg.distributions.uniform_distribution.UniformDistribution)
def test_metrics(self):
with self.test_session():
gan = mock_gan()
loss = BoundaryEquilibriumLoss(gan, loss_config)
d_loss, g_loss = loss.create()
metrics = loss.metrics
self.assertTrue(metrics['k'] != None)
def test_config(self):
with self.test_session():
gan = mock_gan(batch_size=32)
gan.latent = UniformDistribution(gan, {'z':2, 'min': -1, 'max': 1, 'projections':['identity']})
sampler = GridSampler(gan)
self.assertEqual(sampler._sample()['generator'].shape[-1], 1)
def test_adds_supervised_loss(self):
with self.test_session():
gan = mock_gan(y=2)
args = hc.Config({"size": "1", "steps": 1, "method": "train", "save_every": -1, "classloss": True})
cli = hg.CLI(gan, args)
cli.add_supervised_loss()
self.assertEqual(type(cli.gan.loss), MultiComponent)
self.assertEqual(type(cli.gan.loss.components[0]), SupervisedLoss)
def test_create(self):
with self.test_session():
gan = mock_gan()
loss = SupervisedLoss(gan, loss_config)
d_loss, g_loss = loss.create()
d_shape = gan.ops.shape(d_loss)
self.assertEqual(d_shape, [1])
self.assertEqual(g_loss, None)
def test_categories(self):
with self.test_session():
gan = mock_gan()
encoder = CategoryEncoder(gan, config)
gan.encoder = encoder
gan.encoder.create()
gan.create()
self.assertEqual(
gan.ops.shape(encoder.sample)[1] // 2, len(encoder.categories))
def test_config(self):
with self.test_session():
gan = mock_gan(batch_size=32)
gan.latent = UniformDistribution(gan, {'z':2, 'min': -1, 'max': 1, 'projections':['identity']})
gan.latent.create()
gan.create()
sampler = GridSampler(gan)
self.assertEqual(sampler._sample()['generator'].shape[-1], 1)
def test_create(self):
with self.test_session():
gan = mock_gan()
loss = LeastSquaresLoss(gan, loss_config)
d_loss, g_loss = loss.create()
d_shape = gan.ops.shape(d_loss)
g_shape = gan.ops.shape(g_loss)
self.assertEqual(sum(d_shape), 0)
self.assertEqual(sum(g_shape), 0)
def test_output_string(self):
with self.test_session():
gan = mock_gan()
config = {'d_learn_rate': 1e-3, 'g_learn_rate': 1e-3, 'd_trainer': 'rmsprop', 'g_trainer': 'adam'}
trainer = AlternatingTrainer(gan, config)
c = tf.constant(1)
self.assertTrue('d_loss' in trainer.output_string({'d_loss':c}))
self.assertTrue('g_loss' in trainer.output_string({'g_loss':c}))
self.assertEqual(len(trainer.output_variables({'a': c, 'b': c})), 2)
def test_create(self):
with self.test_session():
gan = mock_gan()
loss = LambGanLoss(gan, loss_config)
d_loss, g_loss = loss.create()
d_shape = gan.ops.shape(d_loss)
g_shape = gan.ops.shape(g_loss)
self.assertEqual(d_shape, [])
self.assertEqual(g_shape, [])
def test_create(self):
with self.test_session():
gan = mock_gan()
loss = LeastSquaresLoss(gan, loss_config)
d_loss, g_loss = loss.create()
d_shape = gan.ops.shape(d_loss)
g_shape = gan.ops.shape(g_loss)
self.assertEqual(sum(d_shape), 0)
self.assertEqual(sum(g_shape), 0)
def test_output_string(self):
with self.test_session():
gan = mock_gan()
gan.create()
config = {'d_learn_rate': 1e-3, 'g_learn_rate': 1e-3, 'd_trainer': 'rmsprop', 'g_trainer': 'adam'}
trainer = AlternatingTrainer(gan, config)
c = tf.constant(1)
self.assertTrue('d_loss' in trainer.output_string({'d_loss':c}))
self.assertTrue('g_loss' in trainer.output_string({'g_loss':c}))
self.assertEqual(len(trainer.output_variables({'a': c, 'b': c})), 2)
def test_create(self):
with self.test_session():
gan = mock_gan()
gan.create()
loss = BoundaryEquilibriumLoss(gan, loss_config)
d_loss, g_loss = loss.create()
d_shape = gan.ops.shape(d_loss)
g_shape = gan.ops.shape(g_loss)
self.assertEqual(sum(d_shape), 0)
self.assertEqual(sum(g_shape), 0)
def test_new(self):
with self.test_session():
try:
os.remove('test.json')
except Exception:
pass
gan = mock_gan()
args = hc.Config({"size": "1", "steps": 1, "method": "train", "directory": "test"})
cli = hg.CLI(gan, args)
cli.new()
self.assertTrue(os.path.isfile('test.json'))
def test_config(self):
with self.test_session():
config = {
'd_learn_rate': 1e-3,
'g_learn_rate': 1e-3,
'd_trainer': 'rmsprop',
'g_trainer': 'adam'
}
gan = mock_gan()
trainer = AlternatingTrainer(gan, config)
self.assertEqual(trainer.config.d_learn_rate, 1e-3)
def test_train(self):
with self.test_session():
gan = mock_gan()
args = hc.Config({
"size": "1",
"steps": 1,
"method": "train",
"save_every": -1
})
cli = hg.CLI(gan, args)
cli.train()
self.assertEqual(cli.gan, gan)
def test_create(self):
return None #disabled for now
with self.test_session():
remove_d_config = hg.Configuration.default()
remove_d_config['discriminator'] = None
remove_d_config['loss'] = None
remove_d_config['trainer'] = None
gan = mock_gan(config = remove_d_config)
discriminator = AutoencoderDiscriminator(gan, config)
gan.encoder = gan.create_component(gan.config.encoder)
gan.generator = gan.create_component(gan.config.generator)
self.assertEqual(int(net.get_shape()[1]), 32)
def test_sample(self):
with self.test_session():
gan = mock_gan()
mock_sample = tf.constant(1., shape=[1,1])
multi = MultiComponent(combine='concat',
components=[
MockLoss(gan, sample=mock_sample),
MockLoss(gan, sample=mock_sample)
])
gan.encoder = multi
self.assertEqual(gan.ops.shape(multi.sample), [1,2])
def test_create(self):
return None #disabled for now
with self.test_session():
remove_d_config = hg.Configuration.default()
remove_d_config['discriminator'] = None
remove_d_config['loss'] = None
remove_d_config['trainer'] = None
gan = mock_gan(config=remove_d_config)
discriminator = AutoencoderDiscriminator(gan, config)
gan.encoder = gan.create_component(gan.config.encoder)
gan.generator = gan.create_component(gan.config.generator)
self.assertEqual(int(net.get_shape()[1]), 32)
def test_safe_new(self):
with self.test_session():
try:
os.remove('test.json')
except Exception:
pass
gan = mock_gan()
args = hc.Config({"size": "1", "steps": 1, "method": "train", "directory": "test"})
cli = hg.CLI(gan, args)
cli.new()
with self.assertRaises(ValidationException):
cli.new()
def test_proxy_methods(self):
with self.test_session():
gan = mock_gan()
mock_sample = tf.constant(1., shape=[1, 1])
multi = MultiComponent(combine='concat',
components=[
MockLoss(gan, sample=mock_sample),
MockLoss(gan, sample=mock_sample)
])
multi.proxy()
self.assertEqual(multi.proxy_called, [True, True])
def test_proxy_methods(self):
with self.test_session():
gan = mock_gan()
mock_sample = tf.constant(1., shape=[1,1])
multi = MultiComponent(combine='concat',
components=[
MockLoss(gan, sample=mock_sample),
MockLoss(gan, sample=mock_sample)
])
multi.proxy()
self.assertEqual(multi.proxy_called, [True, True])
def test_sample(self):
with self.test_session():
gan = mock_gan()
mock_sample = tf.constant(1., shape=[1, 1])
multi = MultiComponent(combine='concat',
components=[
MockLoss(gan, sample=mock_sample),
MockLoss(gan, sample=mock_sample)
])
gan.encoder = multi
self.assertEqual(gan.ops.shape(multi.sample), [1, 2])
def test_combine_dict(self):
with self.test_session():
gan = mock_gan()
ops = gan.ops
mock_sample = tf.constant(1., shape=[1])
multi = MultiComponent(combine='add',
components=[
MockLoss(gan, sample={"a": "b"}),
MockLoss(gan, sample={"b": "c"})
])
self.assertEqual(len(multi.sample), 2)
self.assertEqual(multi.sample['a'], 'b')
self.assertEqual(multi.sample['b'], 'c')
def test_sample_loss(self):
with self.test_session():
gan = mock_gan()
ops = gan.ops
mock_sample = tf.constant(1., shape=[1])
multi = MultiComponent(combine='add',
components=[
MockLoss(gan, sample=[mock_sample, None]),
MockLoss(gan, sample=[mock_sample, mock_sample])
])
self.assertEqual(len(multi.sample), 2)
self.assertEqual(ops.shape(multi.sample[0]), [1])
self.assertEqual(ops.shape(multi.sample[1]), [1])
def test_combine_dict(self):
with self.test_session():
gan = mock_gan()
ops = gan.ops
mock_sample = tf.constant(1., shape=[1])
multi = MultiComponent(combine='add',
components=[
MockLoss(gan, sample={"a":"b"}),
MockLoss(gan, sample={"b":"c"})
])
self.assertEqual(len(multi.sample), 2)
self.assertEqual(multi.sample['a'], 'b')
self.assertEqual(multi.sample['b'], 'c')
def test_sample(self):
with self.test_session():
gan = mock_gan()
args = hc.Config({
"size": "1",
"steps": 1,
"method": "train",
"save_every": -1
})
gan.create()
cli = hg.CLI(gan, args)
cli.sample('/tmp/test-sample.png')
self.assertEqual(cli.gan, gan)
def test_adds_supervised_loss(self):
with self.test_session():
gan = mock_gan(y=2)
args = hc.Config({
"size": "1",
"steps": 1,
"method": "train",
"save_every": -1,
"classloss": True
})
cli = hg.CLI(gan, args)
cli.add_supervised_loss()
self.assertEqual(type(cli.gan.loss), MultiComponent)
self.assertEqual(type(cli.gan.loss.components[0]), SupervisedLoss)
def test_new(self):
with self.test_session():
try:
os.remove('test.json')
except Exception:
pass
gan = mock_gan()
args = hc.Config({
"size": "1",
"steps": 1,
"method": "train",
"directory": "test"
})
cli = hg.CLI(gan, args)
cli.new()
self.assertTrue(os.path.isfile('test.json'))
def test_sample_loss(self):
with self.test_session():
gan = mock_gan()
ops = gan.ops
mock_sample = tf.constant(1., shape=[1])
multi = MultiComponent(combine='add',
components=[
MockLoss(gan,
sample=[mock_sample, None]),
MockLoss(
gan,
sample=[mock_sample, mock_sample])
])
self.assertEqual(len(multi.sample), 2)
self.assertEqual(ops.shape(multi.sample[0]), [1])
self.assertEqual(ops.shape(multi.sample[1]), [1])
def test_safe_new(self):
with self.test_session():
try:
os.remove('test.json')
except Exception:
pass
gan = mock_gan()
args = hc.Config({
"size": "1",
"steps": 1,
"method": "train",
"directory": "test"
})
cli = hg.CLI(gan, args)
cli.new()
with self.assertRaises(ValidationException):
cli.new()
def test_config(self):
with self.test_session():
gan = mock_gan()
sampler = BatchSampler(gan)
self.assertEqual(sampler._sample()['generator'].shape[-1], 1)
def test_config(self):
with self.test_session():
loss = LambGanLoss(mock_gan(), loss_config)
self.assertTrue(loss.config.test)
def test_create_component(self):
with self.test_session():
gan = mock_gan()
distribution = gan.create_component(gan.config.latent)
self.assertEqual(type(distribution), hg.distributions.uniform_distribution.UniformDistribution)
def test_config(self):
with self.test_session():
loss = LeastSquaresLoss(mock_gan(), loss_config)
self.assertTrue(loss.config.test)
def test_config(self):
with self.test_session():
loss = BoundaryEquilibriumLoss(mock_gan(), loss_config)
self.assertTrue(loss.config.test)
def test_validate(self):
with self.assertRaises(ValidationException):
gan = mock_gan()
AlternatingTrainer(gan, {})
def test_config(self):
with self.test_session():
config = {'d_learn_rate': 1e-3, 'g_learn_rate': 1e-3, 'd_trainer': 'rmsprop', 'g_trainer': 'adam'}
gan = mock_gan()
trainer = AlternatingTrainer(gan, config)
self.assertEqual(trainer.config.d_learn_rate, 1e-3)
def test_hg_gan(self):
self.assertIs(type(mock_gan()), hg.gans.standard_gan.StandardGAN)
def test_config(self):
with self.test_session():
loss = SupervisedLoss(mock_gan(), loss_config)
self.assertTrue(loss.config.test)
def test_can_create_alphagan(self):
config = hg.Configuration.load('alpha-default.json')
self.assertIs(type(mock_gan(config=config)),
hg.gans.alpha_gan.AlphaGAN)
def test_cli(self):
with self.test_session():
gan = mock_gan()
args = {}
cli = hg.CLI(gan, args)
self.assertEqual(cli.gan, gan)
def test_config(self):
with self.test_session():
return None # disable for now
gan = mock_gan()
discriminator = PyramidDiscriminator(gan, config)
self.assertEqual(discriminator.config.activation, tf.nn.tanh)
def test_can_create_default(self):
config = hg.Configuration.load('default.json')
self.assertIs(type(mock_gan(config=config)), hg.gans.standard_gan.StandardGAN)
def test_config(self):
with self.test_session():
loss = LeastSquaresLoss(mock_gan(), loss_config)
self.assertTrue(loss.config.test)
