CLASStorch.utils.data.TensorDataset(*tensors)

CLASStorch.utils.data.TensorDataset(*tensors)

Dataset wrapping tensors.

Each sample will be retrieved by indexing tensors along the first dimension.

Parameters

*tensors (Tensor) – tensors that have the same size of the first dimension.

 

train_features=torch.Tensor([[1.1,2.1,3.1],[4.1,5.1,6.1],[7.1,8.1,9.1],[10.1,11.1,12.1]])

train_labels=torch.Tensor([[1.1],[2.1],[3.1],[4.1]])

dataset = torch.utils.data.TensorDataset(train_features,

train_labels)

print(dataset)

for i in dataset:

    print(i)

輸出結果：

<torch.utils.data.dataset.TensorDataset object at 0x0000023D5A814B38>
(tensor([1.1000, 2.1000, 3.1000]), tensor([1.1000]))
(tensor([4.1000, 5.1000, 6.1000]), tensor([2.1000]))
(tensor([7.1000, 8.1000, 9.1000]), tensor([3.1000]))
(tensor([10.1000, 11.1000, 12.1000]), tensor([4.1000]))

該函數將行數或列數相同的倆個維度數組進行拼接，在這個代碼中倆個數據按行包裝。

batch_size=2

train_iter = torch.utils.data.DataLoader(dataset,batch_size,shuffle=True)

print(train_iter)

for X,y in train_iter:

    print(X)

    print(y)

<torch.utils.data.dataloader.DataLoader object at 0x0000024E1888B898>

tensor([[1.1000, 2.1000, 3.1000],
[4.1000, 5.1000, 6.1000]])
tensor([[1.1000],
[2.1000]])
tensor([[ 7.1000, 8.1000, 9.1000],
[10.1000, 11.1000, 12.1000]])
tensor([[3.1000],
[4.1000]])

CLASStorch.utils.data.DataLoader，如圖可知DataLoader是將包裝好的n*(特征向量+標簽)分成n/x批，每批(x*特征向量，x*標簽)。該函數返回的是一個迭代對象。batch_size是批大小。