Github user debasish83 commented on the pull request:
https://github.com/apache/spark/pull/1290#issuecomment-82057716
@avulanov could you please point me to a stable branch that I can
experiment with..I am focused on collaborative filtering and implemented
various matrix factorization formulations (quadratic and nonlinear forms):
https://issues.apache.org/jira/browse/SPARK-2426
https://issues.apache.org/jira/browse/SPARK-6323
Sparse Coding and PLSA are very useful for feature extraction but there are
paper where neural nets (autoencoder) have beaten both of them.
I want to start experimenting with the autoencoder variants of neural net.
Specifically I will be focused on these 3 aspects:
1. Distributing the neural net gradient calculation (I think you have
already distributed the gradient calculation which is exactly what I want). If
not we should do it on graphx similar to LDA architecture.
2. A block coordinate descent solver running on Master with BFGS/OWLQN as
the inner solver.
3. Use L1 regularization in place of drop out heuristic to automatically
select interesting features to make the model sparse
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
