about the form of input data to the model

[wowSUNBOY]

Thanks for your code which is really useful.

And I have a question:

During the model definition, the input form should be (batch_size, seq_len, input_size) since the batch_first=True.

But when running the code, the trajectory data are sent into LSTM as the form (seq_len, batch_size, input_size) which is on the opposite.

Such as the 1st trajectory whose length is 192, the data x which is sent to LSTM is in (192, 1, 15).

Can you help me with this problem? Very appreciate.

[jiaxiang-cheng]

Thanks for your code which is really useful.

And I have a question:

During the model definition, the input form should be (batch_size, seq_len, input_size) since the batch_first=True.

But when running the code, the trajectory data are sent into LSTM as the form (seq_len, batch_size, input_size) which is on the opposite.

Such as the 1st trajectory whose length is 192, the data x which is sent to LSTM is in (192, 1, 15).

Can you help me with this problem? Very appreciate.

can you spot exact locations of the issues

[zl13133581232]

您好请问这里没有划分训练测试和验证么

[jiaxiang-cheng]

您好请问这里没有划分训练测试和验证么

您好，训练和测试集是分开的，验证集可根据训练效果自己划分。

[mmm656]

您好，请问为什么在class lstm1的forward里用hn_o+hn_1作为relu层的输入，而不直接用hn_1或output？