def noise_sampler(bs):
return np.random.normal(0.0, 1.0, [bs, 3])
if __name__ == '__main__':
from a_nice_mc.objectives.bayes_logistic_regression.synthetic import Synthetic
from a_nice_mc.models.discriminator import MLPDiscriminator
from a_nice_mc.models.generator import create_nice_network
from a_nice_mc.train.wgan_nll import Trainer
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
energy_fn = Synthetic(batch_size=32)
discriminator = MLPDiscriminator([800, 800, 800])
generator = create_nice_network(3, 10, [
([400], 'v1', False),
([400, 400], 'x1', True),
([400], 'v2', False),
])
trainer = Trainer(generator,
energy_fn,
discriminator,
noise_sampler,
b=16,
m=4,
eta=5)
trainer.train(bootstrap_steps=3000, bootstrap_burn_in=1000)
def noise_sampler(bs):
return np.random.normal(0.0, 1.0, [bs, 750])
if __name__ == '__main__':
from a_nice_mc.objectives.neural_network_regression.boston import Boston
from a_nice_mc.models.discriminator import MLPDiscriminator
from a_nice_mc.models.generator import create_nice_network
from a_nice_mc.train.wgan_nll import Trainer
# os.environ['CUDA_VISIBLE_DEVICES'] = '0'
energy_fn = Boston()
discriminator = MLPDiscriminator([800, 800, 800])
generator = create_nice_network(energy_fn.theta_dim, 50, [
([400], 'v1', False),
([400, 400], 'x1', True),
([400], 'v2', False),
])
trainer = Trainer(generator,
energy_fn,
discriminator,
noise_sampler,
b=16,
m=2)
trainer.train(bootstrap_steps=100,
bootstrap_burn_in=0,
bootstrap_discard_ratio=0.8,
hmc_epochs=1)
