0

所以我有一个我制作的 Dataset 类，它接受一个 3D numpy 数组和形状长度pack_padded_sequence：

class MyDataset(data.Dataset):
     def __init__(self, dataset, data_shape):
          self.dataset = dataset
          self.transform = MyToTensor(data_shape)

我创建了自己的 ToTensor 类：

class MyToTensor(object):
     def __init__(self, data_shape):
          self.data_shape = data_shape
     def __call__(self, data):
          data = torch.from_numpy(data)
          return rnn.pack_padded_sequence(data, lengths=self.data_shape, batch_first=True)

但是由于某种原因，当print(list(MyDataset(dataset, data_shape)))我得到一个正常的张量对象返回而没有删除填充时。

有关我的输入的更多信息，dataset是一个按顺序排列的 3D numpy 数组：batch size, sequence length, features并且 data_shape 是一个列表，其大小batch_size与表示序列长度的数字相匹配。

序列长度也是从最高序列到最低序列大小的顺序

我的输入示例：

[[[0 0.33000001311302185 1]
  [0 0.4300000071525574 1]
  [0 0.3799999952316284 1]
  ...
  [0 0.33000001311302185 1]
  [0 0.28999999165534973 1]
  [0 0.33000001311302185 1]]

 [[6 0.800000011920929 3]
  [5 0.7300000190734863 3]
  [7 0.8199999928474426 3]
  ...
  [4 0.699999988079071 3]
  [5 0.7799999713897705 3]
  [5 0.7799999713897705 3]]

 [[3 1.0 5]
  [3 1.0 5]
  [3 1.0 5]
  ...
  [3 1.0 5]
  [3 1.0 5]
  [3 1.0 5]]

 ...

 [[4.0 0.7599999904632568 3.0]
  [6.0 0.8100000023841858 3.0]
  [6.0 1.0 3.0]
  ...
  [nan nan nan]
  [nan nan nan]
  [nan nan nan]]

 [[8.0 1.0 0.0]
  [8.0 0.9100000262260437 0.0]
  [9.0 1.0 0.0]
  ...
  [nan nan nan]
  [nan nan nan]
  [nan nan nan]]

 [[5.0 1.0 1.0]
  [4.0 1.0 1.0]
  [4.0 1.0 1.0]
  ...
  [nan nan nan]
  [nan nan nan]
  [nan nan nan]]]

以及对应的data_shape：

(235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 235, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 232, 18, 18, 18)
4

0 回答 0