4

I have been looking almost in every resource I can find on the web to see if someone posted an example of Regression using RNN/LSTM in Caffe (here, here, and here). Unfortunately, it looks like such resources don't exist till now. I'm working with Jeff Donahue version on python.

What I'm looking for is something very simple. For example, if you have a 100 data points of (x,y) pairs. How would you go about:

  1. Creating the input matrix.
  2. Creating the contineuation matrix (do we need it or not?)
  3. Creating the target matrix.
  4. How the prototxt file will look like?
  5. Can we extrapolate (do prediction) from using this model?

Also, how the first three items will be if the data is multi dimensional. For example X is a d-dimensional vector and Y is a k-dimensional vector.

Feel free to use your own examples as long as they cover the steps of formatting the data under Python.

I just wanted to note that I also opened a Caffe User's question for this here.

Community
  • 1
  • 1
mxy
  • 331
  • 2
  • 9
  • I think you need to look at this paper "multi-dimensional RNN" https://arxiv.org/pdf/0705.2011v1.pdf – Charles Chow Nov 08 '16 at 07:08
  • Same problem here; was trying to follow [this](http://christopher5106.github.io/deep/learning/2016/06/07/recurrent-neural-net-with-Caffe.html) tutorial with Caffe's master branch implementation (and also asked on the [Google group](https://groups.google.com/forum/#!topic/caffe-users/_kZBdA38nO8)), outcome: Nothing works and I don't know why. Code available in this [Gist](https://gist.github.com/sunsided/f9cda9cfc926436704bab28473ad182c). – sunside Nov 18 '16 at 17:04

0 Answers0