Options

Logging Results of Parallel Operators

dragoljubdragoljub Member Posts: 241 Contributor II
edited November 2018 in Help
Hi Everyone,

I am having problems logging results of operators that run in parallel (across multiple threads). For example, if I use "Loop Parameters (Parallel)" to sweep an SVM parameter and Include a "Log" operator within "Loop Parameters (Parallel)" the logging results returned are just copies of the most reticent values.

If I use 8 threads to optimize 8 combination of SVM parameters (in parallel) and report performance using Log I get is 8 copies of the same result.

How can I get the log operator to return the individual performance vectors of each parameter combination?  ???

Thanks,  ;D
-Gagi

Answers

  • Options
    dragoljubdragoljub Member Posts: 241 Contributor II
    Once again answering my own questions...  :P

    It seems that you can log results of parallel executions to a file however logging to the RM log window does not work. Remember to check the persistent option to save results immediately. Results are also not stored in any particular meaningfully order just in the order of completion.

    Any idea how to get the log to output results to RM GUI correctly and also how about results ordering, does sorting work with parallel operators?  :D
  • Options
    landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531 Unicorn
    Hi,
    if it should work, you at least have to put the log operator into the multi threaded part of the process and access only operators that are in the same scope. Unfortunately there may still remain some thread synchronization issues remaining unsolved that might affect the sorting, but I'm not too familiar with the current status of the extension.

    Greetings,
      Sebastian
  • Options
    dragoljubdragoljub Member Posts: 241 Contributor II
    Yes this indeed the case. Logging from outside the parallel operator only gives you the last result, but logging from within the parallel operator seems to give you one of the parallel results repeated.

    I would say this is a very important thing to handle since parallel evaluation for things like cross validation and parameter search is one of the key things RM offers!  :-\

    -Gagi
Sign In or Register to comment.