pytorch 分批训练

#返回的数据集，既有标号又有数据from torch.utils.data import Dataset, DataLoaderclass LoadDataset(Dataset):def __init__(self, data):self.x = datadef __len__(self):return self.x.shape[0]def __getitem__(self, idx):return torch.from_numpy(np.array(self.x[idx])).float(),\torch.from_numpy(np.array(idx))x_train = np.loadtxt("./data_s/CITE/cite.txt")
dataset = LoadDataset(x_train)

#在模型训练过程中train_loader = DataLoader(dataset, batch_size=256, shuffle=True)  #把打好序号的数据送入DataLoader中，打乱顺序并分批，每批256组数据。for epoch in range(args.pretrain_epoch):for batch_idx, (x, _) in enumerate(train_loader):   #遍历每批数据x = x.to(device)                                #将一批数据送入device中 z, x_bar = model(x)loss = F.mse_loss(x_bar, x)optimizer.zero_grad()loss.backward()optimizer.step()with torch.no_grad():x = torch.Tensor(dataset.x).to(device)          #此处数据一定要从原始的dataset中取，否则还是一批数据，而标签数据是所有的数据，会造成数据数量不匹配# x = dataset.x.to(device)z, x_bar = model(x)loss = F.mse_loss(x_bar, x)print('{} loss: {}'.format(epoch, loss))kmeans = KMeans(n_clusters=args.n_clusters, n_init=20).fit(z.data.cpu().numpy())eva(y_labels, kmeans.labels_, epoch)torch.save(model.state_dict(), args.pretrain_path)

pytorch 分批训练

#返回的数据集，既有标号又有数据from torch.utils.data import Dataset, DataLoaderclass LoadDataset(Dataset):def __init__(self, data):self.x = datadef __len__(self):return self.x.shape[0]def __getitem__(self, idx):return torch.from_numpy(np.array(self.x[idx])).float(),\torch.from_numpy(np.array(idx))x_train = np.loadtxt("./data_s/CITE/cite.txt")
dataset = LoadDataset(x_train)

#在模型训练过程中train_loader = DataLoader(dataset, batch_size=256, shuffle=True)  #把打好序号的数据送入DataLoader中，打乱顺序并分批，每批256组数据。for epoch in range(args.pretrain_epoch):for batch_idx, (x, _) in enumerate(train_loader):   #遍历每批数据x = x.to(device)                                #将一批数据送入device中 z, x_bar = model(x)loss = F.mse_loss(x_bar, x)optimizer.zero_grad()loss.backward()optimizer.step()with torch.no_grad():x = torch.Tensor(dataset.x).to(device)          #此处数据一定要从原始的dataset中取，否则还是一批数据，而标签数据是所有的数据，会造成数据数量不匹配# x = dataset.x.to(device)z, x_bar = model(x)loss = F.mse_loss(x_bar, x)print('{} loss: {}'.format(epoch, loss))kmeans = KMeans(n_clusters=args.n_clusters, n_init=20).fit(z.data.cpu().numpy())eva(y_labels, kmeans.labels_, epoch)torch.save(model.state_dict(), args.pretrain_path)