mboehm7 commented on pull request #1133:
URL: https://github.com/apache/systemds/pull/1133#issuecomment-751864222


   LGTM - welcome to SystemDS and thanks for the patch @ywcb00. During the 
merge, I made a number of smaller corrections - don't worry about them but 
please keep them in mind for the future. 
   
   First, avoid git-merges and an unnecessarily fine-grained commit log as this 
creates lots of conflicts and issues during the final merge (I actually took 
the code, and created a commit from scratch under your name). You can rebase to 
master if necessary. Second, regarding formatting, we use 2-space indentation 
for scripts, but tab-indentation for Java code (your code used 2-spaces there 
too). Similarly, try to avoid long generic variable names (e.g., 
execution_context, matrix_object_1). Third, I fixed the test to avoid 
unnecessary state updates (test_name), use spark execution type for the 
reference runs too, and moved this test to the primitives package (which is 
already part of our github actions). Fourth, I changed the handling of expected 
federated inputs to throw an exception on unsupported combinations to avoid 
silent errors of not producing the output. 
   
   If you want to continue with the quaternary operations, I would recommend to 
(1) handle the remaining loss functions wsloss and wsigmoid, then (2) the 
factor updates wdivmm and wumm. Once we have them all, we can add respective 
algorithm-level tests like ALS-CG. Thanks for your contributions.     
   
   Btw, please add you chosen commit email address (in your commit message) 
with your github handle (e.g., as a secondary email address), otherwise your 
account it not properly linked to the commit and you don't show up as a 
contributor.


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to