def test_mapos(self):
dataset = load.load_nbody_dataset(ncubes=1,
spix=32,
resolution=256,
Mpch=350)
maps = self.list_map_to_test()
X = dataset.get_all_data().flatten()
for name, forward, backward in maps:
print('Test map: {}'.format(name))
x = forward(X)
print(np.sum(np.abs(forward(backward(x)) - x)))
assert (np.sum(np.abs(forward(np.round(backward(x))) - x)) < 1)
params['optimization'] = params_optimization
params['summary_every'] = 500 # Tensorboard summaries every ** iterations
params['print_every'] = 50 # Console summaries every ** iterations
params['save_every'] = 1000 # Save the model every ** iterations
params['summary_dir'] = os.path.join(global_path, name +'_summary/')
params['save_dir'] = os.path.join(global_path, name + '_checkpoints/')
resume, params = utils.test_resume(try_resume, params)
params['Nstats'] = 30
params['Nstats_cubes'] = 10
class CosmoUpscalePatchWGAN(UpscalePatchWGAN, CosmoWGAN):
pass
wgan = UpscaleGANsystem(CosmoUpscalePatchWGAN, params)
dataset = load.load_nbody_dataset(
spix=ns,
scaling=1,
resolution=256,
Mpch=350,
patch=True,
augmentation=True,
forward_map=forward,
is_3d=True)
wgan.train(dataset, resume=resume)
def test_cosmo(self):
forward = fmap.forward
dataset = load.load_nbody_dataset(ncubes=None,
spix=32,
Mpch=350,
forward_map=forward,
patch=True)
it = dataset.iter(10)
print(next(it).shape)
assert (next(it).shape == (10, 32, 32, 4))
del it, dataset
dataset = load.load_nbody_dataset(ncubes=None,
spix=32,
Mpch=350,
forward_map=forward,
patch=True,
is_3d=True)
it = dataset.iter(4)
print(next(it).shape)
assert (next(it).shape == (4, 32, 32, 32, 8))
del it, dataset
dataset = load.load_nbody_dataset(ncubes=None,
spix=32,
Mpch=70,
forward_map=None,
patch=False)
it = dataset.iter(10)
print(next(it).shape)
assert (next(it).shape == (10, 32, 32, 1))
del it, dataset
dataset = load.load_nbody_dataset(ncubes=2,
spix=256,
Mpch=70,
forward_map=forward,
patch=False)
assert (dataset.get_all_data().shape[0] == 256 * 2)
del dataset
dataset = load.load_nbody_dataset(ncubes=2,
spix=128,
Mpch=350,
forward_map=forward,
patch=False)
it = dataset.iter(10)
print(next(it).shape)
assert (next(it).shape == (10, 128, 128, 1))
del it, dataset
dataset1 = load.load_nbody_dataset(ncubes=4,
spix=128,
Mpch=350,
forward_map=forward,
patch=False,
shuffle=False,
augmentation=False,
scaling=2,
is_3d=True)
it1 = dataset1.iter(3)
s1 = next(it1)
del it1, dataset1
dataset2 = load.load_nbody_dataset(ncubes=4,
spix=32,
Mpch=350,
forward_map=forward,
patch=False,
shuffle=False,
augmentation=False,
scaling=8,
is_3d=True)
it2 = dataset2.iter(3)
s2 = next(it2)
del it2, dataset2
np.testing.assert_allclose(np_downsample_3d(s1, 4), s2)
dataset1 = load.load_nbody_dataset(ncubes=2,
spix=128,
Mpch=350,
forward_map=forward,
patch=False,
shuffle=False,
augmentation=False,
scaling=2)
it1 = dataset1.iter(10)
s1 = next(it1)
del it1, dataset1
dataset2 = load.load_nbody_dataset(ncubes=2,
spix=32,
Mpch=350,
forward_map=forward,
patch=False,
shuffle=False,
augmentation=False,
scaling=8)
it2 = dataset2.iter(10)
s2 = next(it2)
del it2, dataset2
np.testing.assert_allclose(np_downsample_2d(s1, 4), s2)
try_resume = True # Try to resume previous simulation
Mpch = 350 # Type of dataset (select 70 or 350)
forward = fmap.stat_forward
backward = fmap.stat_backward
def non_lin(x):
return tf.nn.relu(x)
global_path = '../saved_results/nbody-2d/'
dataset = load.load_nbody_dataset(ncubes=30,
spix=ns,
Mpch=Mpch,
forward_map=forward)
name = 'WGAN{}'.format(ns) + 'test_full_' + '2D'
bn = False
md = 32
params_discriminator = dict()
params_discriminator['stride'] = [1, 2, 2, 2, 1]
params_discriminator['nfilter'] = [md, 2 * md, 4 * md, 2 * md, md]
params_discriminator['shape'] = [[4, 4], [4, 4], [4, 4], [4, 4], [4, 4]]
params_discriminator['batch_norm'] = [bn, bn, bn, bn, bn]
params_discriminator['full'] = []
params_discriminator['minibatch_reg'] = False