def train_gowalla(args): p_str = "running " + args.dataset print(p_str) dataset_base_path = '../../data/Gowalla/gowalla_x0' ##gowalla user_num = 29858 item_num = 40981 factor_num = args.embed_size batch_size = 2048 * 512 top_k = 20 num_negative_test_val = -1 ##all run_id = args.run_id print(run_id) dataset = 'gowalla' path_save_base = './log/' + dataset + '/newloss' + run_id if (os.path.exists(path_save_base)): print('has results save path') else: os.makedirs(path_save_base) result_file = open(path_save_base + '/results.txt', 'w+') #('./log/results_gcmc.txt','w+') copyfile('../../external/LR_gccf/train_gowalla.py', path_save_base + '/train_gowalla' + run_id + '.py') path_save_model_base = './newlossModel/' + dataset + '/s' + run_id if (os.path.exists(path_save_model_base)): print('has model save path') else: os.makedirs(path_save_model_base) training_user_set, training_item_set, training_set_count = np.load( dataset_base_path + '/datanpy/training_set.npy', allow_pickle=True) testing_user_set, testing_item_set, testing_set_count = np.load( dataset_base_path + '/datanpy/testing_set.npy', allow_pickle=True) val_user_set, val_item_set, val_set_count = np.load(dataset_base_path + '/datanpy/val_set.npy', allow_pickle=True) user_rating_set_all = np.load(dataset_base_path + '/datanpy/user_rating_set_all.npy', allow_pickle=True).item() u_d = readD(training_user_set, user_num) i_d = readD(training_item_set, item_num) d_i_train = u_d d_j_train = i_d sparse_u_i = readTrainSparseMatrix(training_user_set, True, u_d, i_d) sparse_i_u = readTrainSparseMatrix(training_item_set, False, u_d, i_d) train_dataset = data_utils.BPRData( train_dict=training_user_set, num_item=item_num, num_ng=5, is_training=True,\ data_set_count=training_set_count,all_rating=user_rating_set_all) train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=2) testing_dataset_loss = data_utils.BPRData( train_dict=testing_user_set, num_item=item_num, num_ng=5, is_training=True,\ data_set_count=testing_set_count,all_rating=user_rating_set_all) testing_loader_loss = DataLoader(testing_dataset_loss, batch_size=batch_size, shuffle=False, num_workers=0) val_dataset_loss = data_utils.BPRData( train_dict=val_user_set, num_item=item_num, num_ng=5, is_training=True,\ data_set_count=val_set_count,all_rating=user_rating_set_all) val_loader_loss = DataLoader(val_dataset_loss, batch_size=batch_size, shuffle=False, num_workers=0) model = BPR(user_num, item_num, factor_num, sparse_u_i, sparse_i_u, d_i_train, d_j_train) model = model.to('cuda') # lr=0.005 optimizer_bpr = torch.optim.Adam(model.parameters(), lr=args.lr) #, betas=(0.5, 0.99)) ########################### TRAINING ##################################### # testing_loader_loss.dataset.ng_sample() print('--------training processing-------') count, best_hr = 0, 0 # epoch=350 for epoch in range(args.epoch): model.train() start_time = time.time() train_loader.dataset.ng_sample() # pdb.set_trace() print('train data of ng_sample is end') # elapsed_time = time.time() - start_time # print(' time:'+str(round(elapsed_time,1))) # start_time = time.time() train_loss_sum = [] train_loss_sum2 = [] for user, item_i, item_j in train_loader: user = user.cuda() item_i = item_i.cuda() item_j = item_j.cuda() model.zero_grad() prediction_i, prediction_j, loss, loss2 = model( user, item_i, item_j) loss.backward() optimizer_bpr.step() count += 1 train_loss_sum.append(loss.item()) train_loss_sum2.append(loss2.item()) # print(count) elapsed_time = time.time() - start_time train_loss = round(np.mean(train_loss_sum[:-1]), 4) #最后一个可能不满足一个batch,所以去掉这样loss就是一致的可以求mean了 train_loss2 = round(np.mean(train_loss_sum2[:-1]), 4) #最后一个可能不满足一个batch,所以去掉这样loss就是一致的可以求mean了 str_print_train = "epoch:" + str(epoch) + ' time:' + str( round(elapsed_time, 1)) + '\t train loss:' + str( train_loss) + "=" + str(train_loss2) + "+" print('--train--', elapsed_time) PATH_model = path_save_model_base + '/epoch' + str(epoch) + '.pt' torch.save(model.state_dict(), PATH_model) model.eval() # ######test and val########### val_loader_loss.dataset.ng_sample() val_loss = evaluate.metrics_loss(model, val_loader_loss, batch_size) # str_print_train+=' val loss:'+str(val_loss) testing_loader_loss.dataset.ng_sample() test_loss = evaluate.metrics_loss(model, testing_loader_loss, batch_size) print(str_print_train + ' val loss:' + str(val_loss) + ' test loss:' + str(test_loss)) result_file.write(str_print_train + ' val loss:' + str(val_loss) + ' test loss:' + str(test_loss)) result_file.write('\n') result_file.flush()
prediction_i = (user * item_i).sum(dim=-1) prediction_j = (user * item_j).sum(dim=-1) # loss=-((rediction_i-prediction_j).sigmoid())**2#self.loss(prediction_i,prediction_j)#.sum() l2_regulization = 0.01 * (user**2 + item_i**2 + item_j**2).sum(dim=-1) # l2_regulization = 0.01*((gcn1_users_embedding**2).sum(dim=-1).mean()+(gcn1_items_embedding**2).sum(dim=-1).mean()) loss2 = -((prediction_i - prediction_j).sigmoid().log().mean()) # loss= loss2 + l2_regulization loss = -( (prediction_i - prediction_j)).sigmoid().log().mean() + l2_regulization.mean() # pdb.set_trace() return prediction_i, prediction_j, loss, loss2 train_dataset = data_utils.BPRData( train_dict=training_user_set, num_item=item_num, num_ng=5, is_training=True,\ data_set_count=training_set_count,all_rating=user_rating_set_all) train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=2) testing_dataset_loss = data_utils.BPRData( train_dict=testing_user_set, num_item=item_num, num_ng=5, is_training=True,\ data_set_count=testing_set_count,all_rating=user_rating_set_all) testing_loader_loss = DataLoader(testing_dataset_loss, batch_size=batch_size, shuffle=False, num_workers=0) val_dataset_loss = data_utils.BPRData(
parser.add_argument("--test_num_ng", type=int, default=99, help="sample part of negative items for testing") parser.add_argument("--out", default=True, help="save model or not") parser.add_argument("--gpu", type=str, default="0", help="gpu card ID") args = parser.parse_args() os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu cudnn.benchmark = True ############################## PREPARE DATASET ########################## train_data, test_data, user_num, item_num, train_mat = data_utils.load_all() # construct the train and test datasets train_dataset = data_utils.BPRData(train_data, item_num, train_mat, args.num_ng, True) test_dataset = data_utils.BPRData(test_data, item_num, train_mat, 0, False) train_loader = data.DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=16) test_loader = data.DataLoader(test_dataset, batch_size=args.test_num_ng + 1, shuffle=False, num_workers=0) ########################### CREATE MODEL ################################# model = model.BPR(user_num, item_num, args.factor_num) model.cuda() optimizer = optim.SGD(model.parameters(), lr=args.lr, weight_decay=args.lamda)
default=50, # the new dataset help="sample part of negative items for testing") parser.add_argument("--out", default=True, help="save model or not") parser.add_argument("--gpu", type=str, default="0", help="gpu card ID") args = parser.parse_args() os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu cudnn.benchmark = True ############################## PREPARE DATASET ########################## train_data, test_data, train_mat, user_num, item_num = data_utils.load_all() # construct the train and test datasets train_dataset = data_utils.BPRData(train_data, user_num, item_num, train_mat, num_ng=args.num_ng, is_training=True) test_dataset = data_utils.BPRData(test_data, user_num, item_num, num_ng=0, is_training=False) train_loader = data.DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=20) # test_loader = data.DataLoader(test_dataset, # batch_size=args.test_num_ng+1, shuffle=False, num_workers=0) test_loader = data.DataLoader(test_dataset, batch_size=1, shuffle=False)