def evaluate(model, data_loader, alpha):
args = parser.arg_parse()
''' set model to evaluate mode '''
#model.eval()
preds = []
gts = [] #ground truth
print('start evaluate')
with torch.no_grad(): # do not need to caculate information for gradient during eval
for idx, (imgs, gt) in enumerate(data_loader):
imgs = imgs.cuda()
pred_class, pred_domain = model(imgs, alpha)
_, pred = torch.max(pred_class, dim = 1)
pred = pred.cpu().numpy().squeeze()
gt = gt.numpy().squeeze()
preds.append(pred)
gts.append(gt)
gts = np.concatenate(gts)
preds = np.concatenate(preds)
np.save(args.save_dir + 'preds.npy', preds)
return accuracy_score(gts, preds) #maybe gts preds#, preds#accuracy_score(gts, preds)
def main():
user = (subprocess.check_output("whoami", shell=True)).rstrip()
args = arg_parse()
if args['which'] == 'launch-ec2':
post_hook(user, args['which'], args['env'], args['role'])
r_code, r_text = launch_ec2(args['instance_type'], args['role'],
args['ip'], args['env'], args['public'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" % r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == "list_lc":
r_code, r_text = list_lc(args['env'])
print r_text
elif args['which'] == "create_lc":
r_code, r_text = create_lc(args['env'], args['instance_type'],
args['role'], args['public'], args['ami'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" % r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == 'bootstrap':
post_hook(user, args['which'], args['env'], args['role'])
r_code, r_text = bootstrap(args['host'], args['role'], args['env'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" % r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == 'codeupdate':
post_hook(user, args['which'], args['env'], args['tag'])
r_code, r_text = codeupdate(args['tag'], args['env'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" % r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == 'adhoc':
r_code, r_text = adhoc(args['host'], args['module'], args['args'],
args['env'])
print r_text
elif args['which'] == 'adhocj':
r_code, r_text = adhocj(args['host'], args['module'], args['args'],
args['env'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" % r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == 'result':
r_code, r_text = result(args['jid'], args['env'])
print r_text
elif args['which'] == 'inventory':
r_code, r_text = ec2list(args['pattern'], args['env'])
print r_text
elif args['which'] == 'version':
bootstrapper_ver = pkg_resources.require("bootstrapper")[0].version
print "Bootstrapper CLI version %s" % bootstrapper_ver
else:
print "invalid command"
def main():
user = (subprocess.check_output("whoami", shell=True)).rstrip()
args = arg_parse()
if args['which'] == 'launch-ec2':
post_hook(user, args['which'], args['env'], args['role'])
r_code,r_text = launch_ec2(args['instance_type'], args['role'], args['ip'], args['env'], args['public'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" %r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == "list_lc":
r_code,r_text = list_lc(args['env'])
print r_text
elif args['which'] == "create_lc":
r_code,r_text = create_lc(args['env'], args['instance_type'], args['role'], args['public'], args['ami'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" %r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == 'bootstrap':
post_hook(user, args['which'], args['env'], args['role'])
r_code,r_text = bootstrap(args['host'], args['role'], args['env'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" %r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == 'codeupdate':
post_hook(user, args['which'], args['env'], args['tag'])
r_code,r_text = codeupdate(args['tag'], args['env'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" %r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == 'adhoc':
r_code,r_text = adhoc(args['host'], args['module'], args['args'], args['env'])
print r_text
elif args['which'] == 'adhocj':
r_code,r_text = adhocj(args['host'], args['module'], args['args'], args['env'])
if r_code == 200:
print "Job Queued Successfully. JobID %s" %r_text
else:
print "Failed to Queue to the Job. Please try again later"
elif args['which'] == 'result':
r_code,r_text = result(args['jid'], args['env'])
print r_text
elif args['which'] == 'inventory':
r_code,r_text = ec2list(args['pattern'], args['env'])
print r_text
elif args['which'] == 'version':
bootstrapper_ver = pkg_resources.require("bootstrapper")[0].version
print "Bootstrapper CLI version %s" %bootstrapper_ver
else:
print "invalid command"
def __init__(self, mode, train_status):
# Load parsed arguments
args = argparser.arg_parse()
mean = argparser.mean
std = argparser.std
# Set directories
self.mode = mode
self.train_status = train_status
self.test_info_dir = args.out_dir_info
self.size = args.size
#distinguish wether model is pre-trained or fine tuned
if train_status == 'pretrain':
self.data_dir = args.data_dir_pre_train
else:
self.data_dir = args.data_dir_fine_tune
#set directories
if self.mode == "train":
gt_dir = os.path.join(self.data_dir, "imgs")
masked_imgs_dir = os.path.join(self.data_dir, "masked_imgs")
masks_dir = os.path.join(self.data_dir, "mask")
elif self.mode == "test" and self.train_status == 'TA':
masked_imgs_dir = os.path.join(args.data_dir_test)
elif self.mode == "test":
masked_imgs_dir = os.path.join(args.data_dir_test, "test")
gt_dir = os.path.join(args.data_dir_test, "test_gt")
else:
print("Invalid mode in dataloader")
sys.exit()
# Get paths to each image. Length = number of images in directory (e.g. 400 for training)
if self.mode == "train":
self.masks = sorted(glob.glob(os.path.join(masks_dir, '*.png')))
self.masked_imgs = sorted(
glob.glob(os.path.join(masked_imgs_dir, '*.jpg')))
elif self.mode == "test":
self.masked_imgs = sorted(
glob.glob(os.path.join(masked_imgs_dir, '*masked.jpg')))
self.masks = sorted(
glob.glob(os.path.join(masked_imgs_dir, '*mask.jpg')))
if not train_status == 'TA':
self.gt_imgs = sorted(glob.glob(os.path.join(gt_dir, '*.jpg')))
# Define transform
self.transformT = transforms.Compose([transforms.ToTensor()])
self.transformN = transforms.Compose([transforms.Normalize(mean, std)])
import argparser as parser
import model_2 as models
import data
import test_source as test
import numpy as np
#from test import evaluate
def save_model(model, save_path):
torch.save(model.state_dict(),save_path)
if __name__=='__main__':
args = parser.arg_parse()
data_mnist =data.Mnist(args, mode='train')
data_SVHN =data.SVHN(args, mode='train')
'''create directory to save trained model and other info'''
if not os.path.exists(args.save_dir):
os.makedirs(args.save_dir)
#''' setup GPU '''
torch.cuda.set_device(args.gpu)
''' setup random seed '''
np.random.seed(args.random_seed)
torch.manual_seed(args.random_seed)
torch.cuda.manual_seed(args.random_seed)
save_model(model, os.path.join(args.model_dir_fine_tune, "Net_finetune_epoch{}.pth.tar".format(epoch+1)))
if epoch > 0:
remove_prev_model(os.path.join(args.model_dir_fine_tune, "Net_finetune_epoch{}.pth.tar".format(epoch)))
print("\n***** Fine-tuning FINISHED *****")
if __name__ == "__main__":
# -------
# General Setup
# -------
# Read input arguments
args = argparser.arg_parse()
# Directory for miscellaneous outputs
if not os.path.exists(args.output_dir):
print("Created directory for outputs: {}".format(args.output_dir))
os.makedirs(args.output_dir)
# Set device
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
# Set seeds for batch shuffling
random.seed(1)
torch.manual_seed(1)
# -------
# Model