jeremiedb commented on issue #7349: time series delay output mxnet
URL: 
https://github.com/apache/incubator-mxnet/issues/7349#issuecomment-321432460
 
 
   symbol.RNN performs the unrolling. 
   Taking a minimal example: 
   
   ```
   data <- mx.symbol.Variable("data")
   rnn.weight <- mx.symbol.Variable("rnn.weight")
   rnn <- mx.symbol.RNN(data = data, num.layers=1, state.size=10)
   regression <- mx.symbol.LinearRegressionOutput(data=rnn, name="final 
regression")
   ```
   Here a sequence of length 51 with batch of 64 and 20 features and sent into 
a RNN which outputs 1 output for each of the 51 elements of fed into the model.
   
   
![graph0](https://user-images.githubusercontent.com/18605903/29151649-bc5f7260-7d50-11e7-8e55-441c416aaf4f.png)
   
   If taking the last element of the sequence, you then feed only the latest 
(51th element) into the regression output. 
   
   
![graph1](https://user-images.githubusercontent.com/18605903/29151650-bf8b7506-7d50-11e7-8a19-71e5c933b704.png)
   
   You might need to hack a little the FeedForward function to account for 
extra parameters required by the RNN cell. The [symbol 
documentation](http://mxnet.io/api/python/symbol.html) should provide the 
required tool to adapt the network according to your needs. 
   
 
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


With regards,
Apache Git Services

Reply via email to