# load Y domain data. Assumed to be a uint8 numpy array
# of shape (n, sz, sz, nc)
def get_y_data():
data_np = np.load(params.y_data_path)
data_np = data_np.transpose((0, 3, 1, 2))
data_np = data_np / 255.0
return data_np
y = get_y_data()
for i in range(y.shape[1]):
y[:, i:i + 1] -= params.mu[i]
y[:, i:i + 1] /= params.sd[i]
rp = np.random.permutation(y.shape[0])[:params.n_examples]
y = y[rp]
net_params = utils.NAMParams(nz=params.nz,
ngf=params.nam_ngf,
mu=params.mu,
sd=params.sd,
force_l2=False)
opt_params = utils.OptParams(lr=params.lr,
batch_size=params.batch_size,
epochs=params.epochs,
decay_epochs=params.decay_epochs,
decay_rate=params.decay_rate,
lr_ratio=params.lr_ratio)
nm = opt_nam.NAMTrainer(netG, params.uncon_shape, net_params)
nm.train_nam(y, opt_params)
try:
from yaml import CLoader as Loader, CDumper as Dumper
except ImportError:
from yaml import Loader, Dumper
with open(args.config, 'r') as f:
params = yaml.load(f, Loader=Loader)
rn = params['name']
train_path = params['train_path']
decay = params['glo']['decay']
total_epoch = params['glo']['total_epoch']
lr = params['glo']['learning_rate']
factor = params['glo']['factor']
nz = params['glo']['nz']
batch_size = params['glo']['batch_size']
do_bn = params['glo']['do_bn']
x = np.load(train_path)
x = x.transpose((0, 3, 1, 2)) / 255.0
glo_params = utils.GLOParams(nz=nz, do_bn=do_bn, force_l2=False)
glo_opt_params = utils.OptParams(lr=lr,
factor=factor,
batch_size=batch_size,
epochs=total_epoch,
decay_epochs=decay,
decay_rate=0.5)
nt = glo.GLOTrainer(x, glo_params, rn)
nt.train_glo(glo_opt_params)