1

Is there any ML framework that readily supports LSTM for multi-dimentionsal data ?

Best would be in Pytorch. But I don't think the official API supports it and I have found an implementation.

Quan Vuong
  • 141
  • 3
  • What's multi-dimensional? What's your loss function in this case? What are you trying to do? – Emre Jun 09 '17 at 05:49
  • In most example, the input to LSTM is a 1d array where each entry is a scalar. I'm trying to input into LSTM an 1d array where each entry is a 2d array. – Quan Vuong Jun 09 '17 at 06:15
  • 2
    @QuanVuong: You can just flatten the 2d array of each entry. Unless you want to process it with something that uses the 2d structure, such as a CNN? In which case you will want to figure out a RNN/CNN hybrid - the way to do this that I have seen, in e.g. video classifier - is to have CNN layers first and RNN replacing fully-connected layer. There are no doubt other architectures. Explaining more about your data and goal should get you a more specific answer. – Neil Slater Jun 09 '17 at 06:45
  • here is a good implementation in tensorflow tf-mdlstm – Cbrom Oct 08 '18 at 12:35

0 Answers0