Baunsgaard commented on PR #1843:
URL: https://github.com/apache/systemds/pull/1843#issuecomment-1714460315

   > Performance Evaluation averaged over 100 runs: <img alt="Screenshot 
2023-09-11 at 19 59 44" width="518" 
src="https://user-images.githubusercontent.com/68165368/267107362-797db0d7-1c44-4475-8d19-94efa751b152.png";>
   > 
   > I observed that the first run takes significantly more time in comparison 
to later runs, although the input matrices are completely randomised.
   
   great! 
   
   There is an option of removing outliers what I typically do is remove the 5% 
best and worst performing runs and then do the average of the remaining. Also 
good practice is to calculate std on the runs and mark it in the graphs with 
+-, this is similarly done without the 5%.
   
   Also for more clarification of where time is spend I would love it if you 
run SystemDS with the -stats  flag.
   this can indicate where we are using time in the DML implementation
   


-- 
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.

To unsubscribe, e-mail: dev-unsubscr...@systemds.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org

Reply via email to