This post introduces how to train the LSTM networks to get correct outputs.

If you use this way, not work.

• Just use the last width, like x[-1, :, :], select the last piece of width.
• It’s normal way. However, sometimes it will NOT WORK.
• 

This way will work.

• Flatten all the output
• 
• It always works.

Have fun.

Besides, I found marimo, which is a replacement for jupyter notebooks.