Wed Aug 15, 2018 1:49 am
Login Register Lost Password? Contact Us


ML with eclserver

Topics related to the set of Machine Learning libraries and Matrix processing algorithms

Thu Apr 30, 2015 4:38 am Change Time Zone

Hi -
Having trouble running some ML (Mat, specifically) with the older eclserver (along with the legacy MySQL code repository).

There *seems* to be a problem reading deeper nested modules.

Should this work?
Will be moving to a folder-based repo eventually - and, therefore, eclccserver - but not yet.

Thanks.
jwilt
 
Posts: 50
Joined: Wed Feb 27, 2013 7:46 pm

Thu Apr 30, 2015 11:35 am Change Time Zone

I don't see how you can fit multiple folder levels into a single folder repository without changing the attributes. Nested modules where the module is a single "file" is all that is possible.

So, you are stuck with needing to modify the current code base to flatten the structure.

The strategy that you are employing, mentioned in ( https://track.hpccsystems.com/browse/ML-250 ), of taking ML.Mat.Add attribute and creating an ML.Mat_Add attribute will indeed be tedious and error prone.

Consider a different approach, one that does not require you to add a bunch of files in the ML repository together into a single large file.

Your classic repository will need a top level folder for each folder in the repository. I suspect that you will need to make the name unique by adding a suffix or prefix. You will then use MODULE atttributes to export aliases. For instance, in the ML folder you will have a single Mat attribute. This attribute will be a MODULE and EXPORT each of the Mat attributes, such as Add, as EXPORT Add(...args...) := Mat_mod.Add.

You will then need to make some changes to the references to use the prefix or suffix. You should be able to use a sed or Perl script to make these changes.

What you gain is that you will no longer need to glob up massive attributes.
john holt
Community Advisory Board Member
Community Advisory Board Member
 
Posts: 22
Joined: Mon Jun 25, 2012 12:43 pm


Return to Machine Learning

Who is online

Users browsing this forum: No registered users and 1 guest