# Dimension of latent vector
zx = 4
zy = 4
maxIter = 100
alpha = np.zeros(maxIter)
for pp in range(maxIter):
alpha[pp] = (2**(maxIter - pp))
# print("alpha")
# print(alpha)
# print(alpha.shape)
# Prepare GAN generator
device = torch.device("cpu")
netG = netG(1, 1, 64, 5, 1)
netG.load_state_dict(torch.load('netG_epoch_5.pth'))
netG.to(device)
netG.eval()
torch.set_grad_enabled(False)
# Z is the observed heads through time
Z = np.loadtxt('ts_ref_gan.txt')
Z = Z.reshape(Z.shape[0], 1)
# Prepare latent_k_array
latent_k_array = torch.rand(nr, 1, zx, zy, device=device) * 2 - 1
print("latent_k_array")
print(latent_k_array)
print(latent_k_array.shape)
plt.matshow(latent_k_array[0][0])
import matplotlib.pyplot as plt
nrow = 129
ncol = 129
nr = 100 # the number of realizations
# Dimension of latent vector
zx = 5
zy = 5
prior_weight = 2
LearningRate = 0.1
# Prepare GAN generator
device = torch.device("cpu")
netG = netG(1, 1, 64, 5, 1)
netG.load_state_dict(torch.load('netG_epoch_10.pth'))
netG.to(device)
netG.eval()
torch.set_grad_enabled(False)
netD = netD(1, 64, 5, 1)
netD.load_state_dict(torch.load('netD_epoch_10.pth'))
netD.to(device)
netD.eval()
torch.set_grad_enabled(False)
Sample_Point = 10
Reference_k = np.loadtxt('Ref_ln_K.txt')
mask_k = np.zeros((nrow, ncol))
Sample = np.zeros((Sample_Point, 2))
mirror=False,
batch_size=batch_size,
n_channel=nc)
# custom weights initialization called on netG and netD
def weights_init(m):
classname = m.__class__.__name__
if classname.find('Conv') != -1:
m.weight.data.normal_(0.0, 0.02)
elif classname.find('BatchNorm') != -1:
m.weight.data.normal_(1.0, 0.02)
m.bias.data.fill_(0)
netG = netG(nc, nz, ngf, gfs, ngpu)
netG.apply(weights_init)
if opt.netG != '':
netG.load_state_dict(torch.load(opt.netG))
print(netG)
netD = netD(nc, ndf, dfs, ngpu=1)
netD.apply(weights_init)
if opt.netD != '':
netD.load_state_dict(torch.load(opt.netD))
print(netD)
criterion = nn.BCELoss()
# Optimizers
# print("real_cpu.shape")
# print(real_cpu.shape)
# print("output.shape")
# print(output.shape)
# print("label.shape")
# print(label.shape)
# print()
errD_real = criterion(output, label)
errD_real.backward()
D_x = output.mean().item()
# train with fake
noise = torch.rand(batch_size, nz, zx, zy, device=device)*2-1
noise_condition, condition_mask = generate_condition(data)
fake = netG(noise, noise_condition)
# print("Ran generator")
# print("noise.shape")
# print(noise.shape)
# print("noise_condition.shape")
# print(noise_condition.shape)
# print("fake.shape")
# print(fake.shape)
# print()
# label.fill_(fake_label)
output = netD(fake.detach())
label = torch.full_like(output, fake_label, device=device) # forcing label to match output count
# print("look here again")
# print(fake.shape)
# print(output.shape)
# noise = torch.rand(batch_size, nz, zx, zy, device=device)*2-1
noise = torch.rand(batch_size, 1, npx, npy, device=device) * 2 - 1
condition = generate_condition(reference_k_array)
# input_matrix.to(device)
print("noise matrix:")
print(noise)
print(noise.shape)
print()
# Turn off gradient calculation
torch.set_grad_enabled(False)
# First run of loop (prepares array for stacking)
# forward run the model
noise = torch.rand(batch_size, 1, npx, npy, device=device) * 2 - 1
output = netG(noise, condition)
print("Output matrix:")
print(output)
print(output.shape)
print()
output = output.cpu()
numpy_output = output.numpy()
numpy_output = numpy_output.squeeze()
print("numpy_output:")
print(numpy_output)
print(numpy_output.shape)
print()
plt.matshow(numpy_output)
# plt.show()
# print("real_cpu.shape")
# print(real_cpu.shape)
# print("output.shape")
# print(output.shape)
# print("label.shape")
# print(label.shape)
# print()
errD_real = criterion(output, label)
errD_real.backward()
D_x = output.mean().item()
# train with fake
noise = torch.rand(batch_size, nz, zx, zy, device=device) * 2 - 1
noise_condition, condition_mask = generate_condition(data)
fake = netG(noise, noise_condition)
# print("Ran generator")
# print("noise.shape")
# print(noise.shape)
# print("noise_condition.shape")
# print(noise_condition.shape)
# print("fake.shape")
# print(fake.shape)
# print()
# label.fill_(fake_label)
output = netD(fake.detach())
label = torch.full_like(
output, fake_label,
device=device) # forcing label to match output count
# print("look here again")
nz = 1 # 'number of non-spatial dimensions in latent space z'
ngf = 64 # 'initial number of filters for dis'
gfs = 5 # 'kernel size for gen'
ngpu = 1 # 'number of GPUs to use'
# Input batch parameters
batch_size = 1
zx = 4
zy = 4
# File directory
outf = './train_data'
epoch = 5
# Load model
netG = netG(nc, nz, ngf, gfs, ngpu)
netG.load_state_dict(torch.load('%s/netG_epoch_%d.pth' % (outf, epoch)))
netG.to(device)
netG.eval()
print(netG)
print()
# Load input matrix
noise = torch.rand(batch_size, nz, zx, zy, device=device)*2-1
# input_matrix.to(device)
print("noise matrix:")
print(noise)
print(noise.shape)
print()
# Turn off gradient calculation