none
guide to creating LSTM with neural network module? RRS feed

  • Question

  • Hey,

    I've been trying to make a rnn using a lstm architecture, and while there seem's to be a good example/tutorial of convolutional nn's using the net# language, I am having trouble getting lstm stuff to work using net# (and I've done this similarily in python, but i would prefer to use the azure modules as they are quicker). Is there a guide for this or plan to release one in the near future?  I am also basing my grasp on this from using theano so I may have to adjust what I am doing.

    Friday, April 24, 2015 6:24 PM

Answers

  • Hello,

    Currently Net# in Azure ML does not support LSTMs. There is a way to create simple RNNs using "shared" keyword in Net# and I'm planning to do a blog post soon which will describe this feature in more details.

    LSTMs are certainly in our feature list but I'm not ready to provide any date on when it actually becomes available.


    Monday, April 27, 2015 7:14 PM

All replies

  • Hello,

    Currently Net# in Azure ML does not support LSTMs. There is a way to create simple RNNs using "shared" keyword in Net# and I'm planning to do a blog post soon which will describe this feature in more details.

    LSTMs are certainly in our feature list but I'm not ready to provide any date on when it actually becomes available.


    Monday, April 27, 2015 7:14 PM
  • Hello, Alexey!

    Did you manage to write a blog post about RNN?
    Is there a way to perform sequence-to-sequence learning in Azure ML?
    Saturday, October 10, 2015 10:13 AM
  • It has been almost a year since this post.

    Has there been any updates on RNN/LSTM and sequence-to-sequence learning in Azure ML?

    Saturday, March 19, 2016 4:16 AM
  • More than four years have passed since this comment; the promised blog post has never been published, no one is answering questions about it, and the people behind Net# seem to have left Microsoft. Is this project abandonware?
    • Edited by xhevahir78 Thursday, September 5, 2019 11:26 PM
    Thursday, September 5, 2019 11:26 PM