opt.n_classes = 10
opt.nz = (100,1,1)
opt.num_disc_iters = 1
opt.checkpoints = [1000, 2000, 5000, 10000, 20000, 40000, 60000, 100000, 200000, 300000, 500000]
basedir = 'oneclass_gans/'
logger = Logger(base_dir=basedir, tag='oneclass_gans')
for digit in range(10):
opt.path = basedir + '{}/'.format(digit)
log_t = Logger()
data = datasets.MNISTDataset(selected=digit)
mydataloader = datasets.MyDataLoader()
data_iter = mydataloader.return_iterator(DataLoader(data, batch_size=opt.batch_size, shuffle=True, num_workers=4), is_cuda=opt.cuda, conditional=opt.conditional, pictures=True)
netG = mnistnet.mnistnet_G(nz=100, ngf=128)
netD = mnistnet.mnistnet_D(nc=1, BN=True, ndf=128)
# netG = mnistnet.mnistnet_G(nz=110)
# netD = mnistnet.mnistnet_D(nc=11)
optimizerD = optim.Adam(netD.parameters(), lr=2e-4, betas=(.5, .999))
optimizerG = optim.Adam(netG.parameters(), lr=2e-4, betas=(.5, .999))
gan1 = gan.GAN(netG=netG, netD=netD, optimizerD=optimizerD, optimizerG=optimizerG, opt=opt)
return [
name for name in os.listdir(dir)
if os.path.isdir(os.path.join(dir, name))
]
# perform some custom validation
if args.TRAIN + args.TEST != 100:
print "error: --train and --test must sum to 100%"
sys.exit(1)
# determine parameters of dataset
if args.DATASET == "feret":
dataset = datasets.FERETDataset()
elif args.DATASET == "mnist":
dataset = datasets.MNISTDataset()
elif args.DATASET == "orl":
dataset = datasets.ORLDataset()
elif args.DATASET == "gtex":
subs = get_sub_dirs('datasets/GTEx_Data')
subs.sort()
dataset = datasets.GTEXDataset(subs)
elif args.DATASET == "gtex_30":
subs = get_sub_dirs('datasets/GTEx_Data_30')
subs.sort()
dataset = datasets.GTEXDataset30(subs)
elif args.DATASET == "fctl":
subs = get_sub_dirs('datasets/fctl')
dataset = datasets.FCTLDataset(subs)
# initialize the training set and test set
# opt.path = 'test_GAN_GANc2st/'.format(i)
opt.path = 'test_GAN_GANc2st/'
writer = SummaryWriter(opt.path)
opt.checkpoints = [
1000, 2000, 5000, 10000, 20000, 40000, 60000, 100000, 200000, 300000,
500000
]
opt.checkpoints = [1000, 2000, 4000, 8000, 12000, 16000, 20000]
logger = Logger(base_dir=opt.path, tag=str(i))
for checkpoint in opt.checkpoints:
real_dataset = datasets.MNISTDataset(selected=i, train=False)
# print(len(real_dataset))
# real_dataset = datasets.LINDataset(protein='Arp3', transform=transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
# basedir='/home/ubuntu/LIN/LIN_Normalized_WT_size-48-80_train/')
log_t = Logger()
loss, roc = c2st(mnistnet_G(nc=1),
'oneclass_gans/{}/gen_{}.pth'.format(i, checkpoint),
mnistnet_D(nc=1),
gan.GAN,
opt,
real_dataset,
logger=log_t)