def main():
flow = build_mera()
last_epoch = utils.get_last_checkpoint_step()
utils.load_checkpoint(last_epoch, flow)
flow.train(False)
shape = (16, args.nchannels, args.L, args.L)
prior_low = Laplace(torch.tensor(0.), torch.tensor(T_low / sqrt(2)))
z = prior_low.sample(shape)
prior_high = Laplace(torch.tensor(0.), torch.tensor(T_high / sqrt(2)))
z_high = prior_high.sample(shape)
k = 2**level_cutoff
z[:, :, ::k, ::k] = z_high[:, :, ::k, ::k]
z = z.to(args.device)
with torch.no_grad():
x, _ = flow.inverse(z)
samples = x.permute(0, 2, 3, 1).detach().cpu().numpy()
samples = 1 / (1 + np.exp(-samples))
fig, axes = plt.subplots(4, 4, figsize=(4, 4), sharex=True, sharey=True)
for i in range(4):
for j in range(4):
ax = axes[i, j]
ax.imshow(samples[j * 4 + i])
ax.axis('off')
plt.tight_layout()
plt.savefig('./mix_T.pdf', bbox_inches='tight')
def test5():
"""
Laplacian distribution is a continuous probability distribution
parameterized by loc and scale
:return:
"""
from torch.distributions.laplace import Laplace
dist = Laplace(torch.tensor([10.0]), torch.tensor([0.99]))
dist.sample() # >>> tensor([10.2167])
def addnoise(self, x):
"""Adds Laplacian noise to histogram of counts
Args:
counts[torch tensor]: Histogram counts
epsilon[integer]:Amount of Noise
Returns:
counts[torch tensor]: Noisy histogram of counts
"""
m = Laplace(torch.tensor([0.0]), torch.tensor([self.epsilon]))
count = x + m.sample()
return count
def sample_laplace(size, gpu=None):
m = Laplace(torch.tensor([0.0]), torch.tensor([1.0]))
y = m.sample(sample_shape=torch.Size([size[0], size[1], size[2]
])).float().squeeze(3)
y = y if gpu is None else y.cuda(gpu)
return y
def addnoise(x):
m = Laplace(torch.tensor([0.0]), torch.tensor([0.5]))
count = x + m.sample().cuda()
return count