尝试理解 Pytorch 的 LSTM 实现

我有一个包含 1000 个示例的数据集，其中每个示例都有5特征（a、b、c、d、e）。我想喂7LSTM 的示例，以便它预测第 8 天的特征 (a)。

阅读 nn.LSTM() 的 Pytorchs 文档，我得出以下结论：

input_size = 5
hidden_size = 10
num_layers = 1
output_size = 1

lstm = nn.LSTM(input_size, hidden_size, num_layers)
fc = nn.Linear(hidden_size, output_size)

out, hidden = lstm(X)  # Where X's shape is ([7,1,5])
output = fc(out[-1])

output  # output's shape is ([7,1])

根据文档：

nn.LSTM 的输入是“输入形状（seq_len、批次、input_size

当我将您的代码扩展为完整的示例时——我还添加了一些可能有帮助的注释——我得到以下信息：

import torch
import torch.nn as nn

input_size = 5
hidden_size = 10
num_layers = 1
output_size = 1

lstm = nn.LSTM(input_size, hidden_size, num_layers)
fc = nn.Linear(hidden_size, output_size)

X = [
    [[1,2,3,4,5]],
    [[1,2,3,4,5]],
    [[1,2,3,4,5]],
    [[1,2,3,4,5]],
    [[1,2,3,4,5]],
    [[1,2,3,4,5]],
    [[1,2,3,4,5]],
]

X = torch.tensor(X, dtype=torch.float32)

print(X.shape)         # (seq_len, batch_size, input_size) = (7, 1, 5)
out, hidden = lstm(X)  # Where X's shape is ([7,1,5])
print(out.shape)       # (seq_len, batch_size, hidden_size) = (7, 1, 10)
out = out[-1]          # Get output of last step
print(out.shape)       # (batch, hidden_size) = (1, 10)
out = fc(out)          # Push through linear layer
print(out.shape)       # (batch_size, output_size) = (1, 1)

这对我来说很有意义，考虑到你batch_size = 1 and output_size = 1（我假设，你正在做回归）。我不知道你在哪里output.shape = (7, 1)来自。

您确定您的X有正确的尺寸吗？你创建了吗nn.LSTM也许与batch_first=True？有很多小东西可以潜入。