[mlpack-svn] [MLPACK] #345: Sparse Autoencoder Module

MLPACK Trac trac at coffeetalk-1.cc.gatech.edu
Thu Apr 17 01:06:50 EDT 2014


#345: Sparse Autoencoder Module
----------------------------+-----------------------------------------------
  Reporter:  siddharth.950  |        Owner:     
      Type:  enhancement    |       Status:  new
  Priority:  major          |    Milestone:     
 Component:  mlpack         |   Resolution:     
  Keywords:                 |     Blocking:     
Blocked By:                 |  
----------------------------+-----------------------------------------------

Comment (by siddharth.950):

 Well, the answer to the first question would be no. Essentially, we are
 trying to fit the data into a smaller number of features, and
 reconstructing the data using those features. If the number of units in
 L_3 and L_1 are not the same, it won't do that (I'm not sure what it will
 do), and also the definition of the cost function will not make sense. We
 could take visibleSize from the data's dimensionality, but I think the
 current implementation would be more useful for debugging. We could add a
 check whether the data that is being passed has the correct dimensions.
 The data matrix is expected to have the examples as columns, but if the
 user passes the transpose of the data matrix, what you are suggesting will
 still work but it won't give the output that the user is expecting. I hope
 that makes sense.

 Thanks for adding it to the trunk! Feels good to have my name as an
 author.

-- 
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/345#comment:13>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.


More information about the mlpack-svn mailing list