if __name__ == '__main__':
torch.cuda.set_device(6)
i_genre = np.load('/home2/zh/data/ml-1m/i_genre.npy')
i_other = np.load('/home2/zh/data/ml-1m/i_other.npy')
ui = np.load('/home2/zh/data/ml-1m/ui.npy')
x_genre = np.load('/home2/zh/data/ml-1m/x_genre.npy')
x_other = np.load('/home2/zh/data/ml-1m/x_other.npy')
y = np.load('/home2/zh/data/ml-1m/y.npy')
config = {'num_embeddings': 3529,
'num_ctns': 1,
'fieldnum': 7,
'embedding_dim': 32,
'headnum': 8,
'attention_dim': 32,
'seed': 81192}
setup_seed(config['seed'])
model = AutoInt(config).cuda().double()
user_set = np.array(list(set(ui[:,0])))
usernum = len(user_set)
perm = random.permutation(usernum)
train_usernum = int(0.8*usernum)
valid_usernum = int(0.9*usernum)
idx0 = np.where(np.isin(ui[:,0], user_set[perm[0:train_usernum]]))[0]
idx = idx0[random.permutation(idx0.shape[0])[:3200]]
onehot_i = torch.tensor(i_other[idx,:-1]).cuda()
onehot_x = torch.tensor(x_other[idx,:-1]).cuda()
multihot_i = torch.tensor(i_genre[idx]).cuda()
multihot_x = torch.tensor(x_genre[idx]).cuda()
multihot_list = [(multihot_i, multihot_x)]
ctns = torch.tensor(x_other[idx,-1:]).cuda()
# label = torch.tensor(y[idx]).cuda().double()
print(f'embedding matrix shape: {pretrained_embeddings.shape}')
print(f'relatedness matrix shape: {related_embeddings.shape}')
return train_dl, valid_dl, pretrained_embeddings, related_embeddings
def get_res(config, train_dl, valid_dl, pretrained_embeddings, related_embeddings):
model = LSTMClassifier(config, pretrained_embeddings, related_embeddings)
model.cuda()
top5, top1 = util.train_model(model, train_dl, valid_dl, config)
del model
return top5, top1
if __name__ == "__main__":
util.setup_seed(6)
parser = argparse.ArgumentParser(description='Knowledge in Labels Project')
parser.add_argument('-d', '--data', help='data name', default='imdb',
choices=['agnews', 'imdb', 'newsgroup'])
parser.add_argument('-g', '--gpu', help='gpu id', type=int, default=0)
args = parser.parse_args()
with open('settings.json', 'r', encoding='utf-8') as f:
settings = json.load(f)
config = settings["lstm"][args.data]
config["epochs"] = 20
config["embed_dropout_prob"] = 0.2
config["vocab_size"] = None
config["data_name"] = args.data
config["embed_dim"] = 300
torch.cuda.set_device(args.gpu)
if args.cfg_file is not None:
cfg_from_file(args.cfg_file)
if args.gpu_id != '-1':
cfg.GPU_ID = args.gpu_id
else:
cfg.CUDA = False
if args.data_dir != '':
cfg.DATA_DIR = args.data_dir
print('Using config')
pprint.pprint(cfg)
if args.manualSeed is None:
args.manualSeed = random.randint(1, 10000)
setup_seed(args.manualSeed)
now = datetime.datetime.now(dateutil.tz.tzlocal())
timestamp = now.strftime('%Y_%m_%d_%H_%M_%S')
output_dir = f'./output/{cfg.Data.dataset}_{cfg.config_name}_{timestamp}'
num_gpu = len(cfg.GPU_ID.split(','))
data = Data()
dataset = (data.trainloader, data.validloader, data.testloader)
algo = Trainer(output_dir, dataset)
start_t = time.time()
algo.train()
end_t = time.time()
print(f'Total time for training:{end_t - start_t}')
with torch.no_grad():
pred, loss = model(task_data, label, spt_idx, qry_idx, phase)
preds.append(pred)
losses.append(loss)
labels.append(label[qry_idx])
auc = roc_auc_score(np.array(torch.cat(labels).detach().cpu()),
np.array(torch.cat(preds).detach().cpu()))
loss = torch.stack(losses).mean().item()
model.train()
return auc, loss
if __name__ == '__main__':
args = parse_args()
torch.cuda.set_device(args.gpunum)
setup_seed(args.seed)
dataloader = MyDataLoader(args)
model = MetaLearner(args).cuda().double()
optimizer = optim.SGD([{
'params': model.net.parameters(),
'lr': args.learning_rate[0]
}, {
'params':
list(model.parameters())[len(list(model.net.parameters())):],
'lr':
args.learning_rate[1]
}])
traucs, vaaucs, trloss, valoss = [], [], [], []
labels_chkp, preds_chkp, losses_chkp = [], [], []
maxauc = 0
test_auc, test_loss = 0, 0