def test_1d(self):
bn = False
params = dict()
params['optimization'] = dict()
params['optimization']['epoch'] = 1
params['summary_every'] = 4
params['save_every'] = 5
params['print_every'] = 3
params['net'] = dict()
params['net']['shape'] = [16, 1] # Shape of the image
params['net']['generator'] = dict()
params['net']['generator']['latent_dim'] = 100
params['net']['generator']['full'] = [2 * 8]
params['net']['generator']['nfilter'] = [2, 32, 32, 1]
params['net']['generator']['batch_norm'] = [bn, bn, bn]
params['net']['generator']['shape'] = [[5], [5], [5], [5]]
params['net']['generator']['stride'] = [1, 2, 1, 1]
params['net']['generator']['data_size'] = 1
params['net']['discriminator'] = dict()
params['net']['discriminator']['full'] = [32]
params['net']['discriminator']['nfilter'] = [16, 32, 32, 32]
params['net']['discriminator']['batch_norm'] = [bn, bn, bn, bn]
params['net']['discriminator']['shape'] = [[5], [5], [5], [3]]
params['net']['discriminator']['stride'] = [2, 2, 2, 1]
params['net']['discriminator']['data_size'] = 1
X = np.random.rand(101, 16)
dataset = Dataset(X)
wgan = GANsystem(WGAN, params)
wgan.train(dataset)
img = wgan.generate(2)
assert (len(img) == 2)
assert (img.shape[1:] == (16, 1))
img = wgan.generate(500)
assert (len(img) == 500)
def test_generate(self):
params = dict()
params['optimization'] = dict()
params['optimization']['epoch'] = 1
params['summary_every'] = 4
params['save_every'] = 5
params['print_every'] = 3
params['net']= dict()
params['net']['generator'] = dict()
params['net']['generator']['latent_dim'] = 5
X = np.random.rand(101,16,16)
dataset = Dataset(X)
wgan = GANsystem(WGAN, params)
wgan.train(dataset)
img = wgan.generate(2)
img = wgan.generate(500)
assert(len(img)==500)
z = np.random.randn(133,5)
img1 = wgan.generate(z=z)
img2 = np.zeros(shape=img1.shape)
for i in range(133):
img2[i] = wgan.generate(z=np.reshape(z[i], [1,5]))
np.testing.assert_almost_equal(img1, img2, decimal=6)
def test_gansystem(self):
params = dict()
params['optimization'] = dict()
params['optimization']['epoch'] = 1
params['summary_every'] = 4
params['save_every'] = 5
params['print_every'] = 3
X = np.random.rand(101,16,16)
dataset = Dataset(X)
wgan = GANsystem(WGAN, params)
wgan.train(dataset)
img = wgan.generate(2)
img = wgan.generate(500)
assert(len(img)==500)
def test_patchupscalegan1dborder(self):
bn = False
params = dict()
params['optimization'] = dict()
params['optimization']['epoch'] = 1
params['summary_every'] = 4
params['save_every'] = 5
params['print_every'] = 3
params['net'] = dict()
params['net']['shape'] = [8, 2]
params['net']['generator'] = dict()
params['net']['generator']['latent_dim'] = 8
params['net']['generator']['full'] = [16]
params['net']['generator']['nfilter'] = [8, 32, 1]
params['net']['generator']['batch_norm'] = [bn, bn]
params['net']['generator']['shape'] = [[3], [3], [3]]
params['net']['generator']['stride'] = [1, 1, 1]
params['net']['generator']['data_size'] = 1
params['net']['generator']['borders'] = dict()
params['net']['generator']['borders']['width_full'] = None
params['net']['generator']['borders']['nfilter'] = [4, 1]
params['net']['generator']['borders']['batch_norm'] = [bn, bn]
params['net']['generator']['borders']['shape'] = [[5], [3]]
params['net']['generator']['borders']['stride'] = [2, 2]
params['net']['generator']['borders']['data_size'] = 1
params['net']['generator']['borders']['width_full'] = 2
params['net']['discriminator'] = dict()
params['net']['discriminator']['full'] = [32]
params['net']['discriminator']['nfilter'] = [16, 32]
params['net']['discriminator']['batch_norm'] = [bn, bn]
params['net']['discriminator']['shape'] = [[5], [3]]
params['net']['discriminator']['stride'] = [2, 2]
params['net']['discriminator']['data_size'] = 1
params['net']['upscaling'] = 2
X = np.random.rand(101, 8, 2)
dataset = Dataset(X)
wgan = GANsystem(UpscalePatchWGAN, params)
wgan.train(dataset)
borders = np.random.rand(500, 8, 1)
X_smooth = np.random.rand(500, 8, 1)
print(borders[:2].shape)
img = wgan.generate(N=2, X_smooth=X_smooth[:2], borders=borders[:2])
assert (len(img) == 2)
assert (img.shape[1:] == (8, 1))
img = wgan.generate(N=500,
X_smooth=X_smooth[:500],
borders=borders[:500])
assert (len(img) == 500)
def test_patchupscalegan3d(self):
bn = False
params = dict()
params['optimization'] = dict()
params['optimization']['epoch'] = 1
params['summary_every'] = 4
params['save_every'] = 5
params['print_every'] = 3
params['net'] = dict()
params['net']['shape'] = [8, 8, 8, 8]
params['net']['generator'] = dict()
params['net']['generator']['latent_dim'] = 8 * 8 * 8
params['net']['generator']['full'] = []
params['net']['generator']['nfilter'] = [8, 32, 1]
params['net']['generator']['batch_norm'] = [bn, bn]
params['net']['generator']['shape'] = [[3, 3, 3], [3, 3, 3], [3, 3, 3]]
params['net']['generator']['stride'] = [1, 1, 1]
params['net']['generator']['data_size'] = 3
params['net']['discriminator'] = dict()
params['net']['discriminator']['full'] = [32]
params['net']['discriminator']['nfilter'] = [16, 32]
params['net']['discriminator']['batch_norm'] = [bn, bn]
params['net']['discriminator']['shape'] = [[3, 3, 3], [3, 3, 3]]
params['net']['discriminator']['stride'] = [2, 2]
params['net']['discriminator']['data_size'] = 3
params['net']['upsampling'] = 2
X = np.random.rand(101, 8, 8, 8, 8)
dataset = Dataset(X)
class UgradedGAN(UpscalePatchWGAN, CosmoWGAN):
pass
wgan = GANsystem(UgradedGAN, params)
wgan.train(dataset)
borders = np.random.rand(500, 8, 8, 8, 7)
X_down = np.random.rand(500, 4, 4, 4, 1)
img = wgan.generate(N=2, X_down=X_down[:2], borders=borders[:2])
assert (len(img) == 2)
assert (img.shape[1:] == (8, 8, 8, 1))
img = wgan.generate(N=500, X_down=X_down[:500], borders=borders[:500])
assert (len(img) == 500)
def test_lapgan2d(self):
bn = False
params = dict()
params['optimization'] = dict()
params['optimization']['epoch'] = 1
params['summary_every'] = 4
params['save_every'] = 5
params['print_every'] = 3
params['net'] = dict()
params['net']['shape'] = [16, 16, 1]
params['net']['generator'] = dict()
params['net']['generator']['latent_dim'] = 16 * 16
params['net']['generator']['full'] = []
params['net']['generator']['nfilter'] = [8, 32, 1]
params['net']['generator']['batch_norm'] = [bn, bn]
params['net']['generator']['shape'] = [[5, 5], [5, 5], [5, 5]]
params['net']['generator']['stride'] = [1, 1, 1]
params['net']['generator']['data_size'] = 2
params['net']['discriminator'] = dict()
params['net']['discriminator']['full'] = [32]
params['net']['discriminator']['nfilter'] = [16, 32]
params['net']['discriminator']['batch_norm'] = [bn, bn]
params['net']['discriminator']['shape'] = [[5, 5], [3, 3]]
params['net']['discriminator']['stride'] = [2, 2]
params['net']['discriminator']['data_size'] = 2
params['net']['upscaling'] = 2
X = np.random.rand(101, 16, 16)
dataset = Dataset(X)
wgan = GANsystem(LapWGAN, params)
wgan.train(dataset)
X_down = np.random.rand(500, 8, 8, 1)
img = wgan.generate(N=2, X_down=X_down[:2])
assert (len(img) == 2)
assert (img.shape[1:] == (16, 16, 1))
img = wgan.generate(N=500, X_down=X_down[:500])
assert (len(img) == 500)
wgan = GANsystem(InpaintingGAN, params)
# Generate new samples
print('Generate new samples')
real_signals = dataset.get_samples(N=N_f)
if model == 'extend':
border1 = real_signals[:, signal_split[0]:(signal_split[0] +
signal_split[1])]
border2 = real_signals[:, -(signal_split[3] +
signal_split[4]):-signal_split[4]]
border3 = real_signals[:, :(signal_split[0] + signal_split[1])]
border4 = real_signals[:, -(signal_split[3] + signal_split[4]):]
borders1 = np.stack([border1, border2], axis=2)
borders2 = np.stack([border3, border4], axis=2)
fake_signals = np.squeeze(wgan.generate(N=N_f,
borders1=borders1,
borders2=borders2)[1],
axis=2)
elif model == 'basic':
border1 = real_signals[:, :signal_split[0]]
border2 = real_signals[:, -signal_split[2]:]
borders = np.stack([border1, border2], axis=2)
fake_signals = np.squeeze(wgan.generate(N=N_f, borders=borders))
# =============================================================================
# import ltfatpy
# from ltfatpy import plotdgtreal
# def plot_sgram(signal, a = 256, M = 512, g='itersine', dynrange=80, **kwargs):
# c = ltfatpy.gabor.dgtreal.dgtreal(signal, g, a, M)[0]
# return plotdgtreal(c, a, M, dynrange=dynrange,**kwargs)
# =============================================================================
wgan = GANsystem(ModSpectrogramGAN, params)
nsamples = 256
nlatent = 100
def clip_dist2(nsamples, nlatent, m=2.5):
shape = [nsamples, nlatent]
z = np.random.randn(*shape)
support = np.logical_or(z<-m, z>m)
while np.sum(support):
z[support] = np.random.randn(*shape)[support]
support = np.logical_or(z<-m, z>m)
return z
d2 = clip_dist2(nsamples, nlatent)
np.max(d2), np.min(d2)
real_signals = dataset.get_samples(N=64)
print(real_signals.shape)
fake_signals = np.squeeze(wgan.generate(N=nsamples, z=d2))
print(real_signals.max())
print(real_signals.min())
print(real_signals.mean())
print(fake_signals.max())
print(fake_signals.min())
print(fake_signals.mean())
import scipy.io
scipy.io.savemat('commands_listen.mat', {"original": real_signals, "generated": fake_signals})
def test_conditionalwgan(self):
bn = False
ns = 16
# Parameters for the generator
params_generator = dict()
params_generator['latent_dim'] = 126
params_generator['stride'] = [1, 2, 1, 1]
params_generator['nfilter'] = [32, 64, 32, 1]
params_generator['shape'] = [[5, 5], [5, 5], [5, 5], [5, 5]]
params_generator['batch_norm'] = [bn, bn, bn]
params_generator['full'] = [8 * 8 * 32]
params_generator['non_lin'] = tf.nn.relu
params_generator['in_conv_shape'] = [8, 8]
# Parameters for the discriminator
params_discriminator = dict()
params_discriminator['stride'] = [2, 2, 1]
params_discriminator['nfilter'] = [16, 64, 32]
params_discriminator['shape'] = [[5, 5], [5, 5], [3, 3]]
params_discriminator['batch_norm'] = [bn, bn, bn]
params_discriminator['full'] = [512, 128, 32]
params_discriminator['minibatch_reg'] = False
# Optimization parameters
d_opt = dict()
d_opt['optimizer'] = "rmsprop"
d_opt['learning_rate'] = 3e-5
params_optimization = dict()
params_optimization['n_critic'] = 5
params_optimization['batch_size'] = 8
params_optimization['epoch'] = 1
# all parameters
params = dict()
params['net'] = dict() # All the parameters for the model
params['net']['generator'] = params_generator
params['net']['discriminator'] = params_discriminator
params['net']['shape'] = [ns, ns, 1] # Shape of the image
params['net']['gamma_gp'] = 10 # Gradient penalty
# Conditional params
params['net']['prior_normalization'] = False
params['net']['cond_params'] = 2
params['net']['init_range'] = [[0, 1], [0, 1]]
params['net']['prior_distribution'] = "gaussian_length"
params['net']['final_range'] = [
0.1 * np.sqrt(params_generator['latent_dim']),
1 * np.sqrt(params_generator['latent_dim'])
]
params['optimization'] = params_optimization
params['summary_every'] = 4
params['save_every'] = 5
params['print_every'] = 3
X = np.random.rand(101, ns, ns)
parameters = np.random.rand(101, 2)
dataset = Dataset_parameters(X, parameters)
wgan = GANsystem(ConditionalParamWGAN, params)
wgan.train(dataset)
img = wgan.generate(N=16,
**{
'z':
wgan.net.sample_latent(bs=16,
params=np.random.rand(
16, 2))
})
assert (img.shape[0] == 16 and img.shape[1] == ns
and img.shape[2] == ns and img.shape[3] == 1)