fid = open('parameters', 'wb+')
for param in module.parameters():
b = param.data.numpy()
fid.write(b)
fid.close()
if use_gpu:
module.cuda()
module = nn.DataParallel(module, gpu)
for stage in ([0] * 1):
# for epoch in range(1):
for phase in ["valid"]:
print("Testing...")
module.train(False)
for param in module.parameters():
param.requires_grad_(False)
running_dist = 0.
for batch, data in enumerate(dataloader[phase], 1):
x, t, idx = data
if use_gpu:
x = x.cuda()
t = t.cuda()
batch_size = 32
bs = 4
for i in range(0, batch_size, bs):
#xm = x[i:i + bs, :, 3:(height-3), 3:(width-3)]
xm = mirror_padding(x[i:i + bs], 21, True)
iterations = 0
for stage in ([0]):
if stage == 0:
rg = range(0, 30000) #700
lr = 0.0003 #* pow(0.9997,2100)
else:
rg = range(2)
lr = 0.0003
for epoch in rg:
lr *= 0.9997
print("\nEpoch {:d}".format(epoch))
for phase in ["train", "valid"]:
if phase == "train":
print("Training...")
module.train(True)
for param in module.parameters():
param.requires_grad_(True)
for param_g in optimizer.param_groups:
param_g['lr'] = lr
else:
print("Validing...")
module.train(False)
for param in module.parameters():
param.requires_grad_(False)
running_dist = 0.
for batch, data in enumerate(dataloader[phase], 1):
x, t, idx = data
if use_gpu:
x = x.cuda()
