RNG behavior across platforms

User8259User8259 Member, University Professor Posts: 8 University Professor
edited September 5 in Help
Can the Global or Local PRNGS in RMS produce different results on different machines all else being the same including the starting seed value and the process? 

Best Answer

  • MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,453 RM Data Scientist
    Solution Accepted
    Hey,
    since RM uses Java's Random generator there should not be any difference across machines if you set all random seeds to a fixed one. I think the only case where it might differ is 32bit vs 64bit, but this should nowadays only be a thing for edge devices.

    Best,
    Martin
    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
    User8259

Answers

  • User8259User8259 Member, University Professor Posts: 8 University Professor
    Thx Martin. Have been observing differences between my laptop, the classroom podium machines (both 64-bit) and some student machines (varies) and have not been able to understand why. I wondered if the PRNG also uses some hardware characteristic like the motherboard serial numbers during seeding that could explain the observed differences (e.g, during sampling, data partitioning). 
  • MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,453 RM Data Scientist
    Hi,
    are you sure you set all the random seeds in all operators to a defined static seed? By default thats not the case.
    BR,
    Martin
    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
  • User8259User8259 Member, University Professor Posts: 8 University Professor
    Hi Martin: I am pretty sure. We all (self and students) picked specific local seed values and some reported results different from the rest of the class. Last semester, I noticed differences between results on my home laptop and results on the podium machine with identical, to the best of my knowledge, process setups. I assumed that RMS was acting up on the podium (as sometimes Excel and Tableau also do) and college tech support opined that it perhaps was because I am not an administrator on the classroom podium machines. But I wondered why. This semester, I am polling students repeatedly to find out if anybody sees results different from the others when we use the Local PRNG. I will see if I can pinpoint the problem as occurring without plausible explanation (e.g., did the student do something different that caused this?) and report back to you. We just got started on RMS and I'll know more as the semester progresses and we get started on projects. Thanks much for the feedback -- I'll be sure to track this more carefully.
Sign In or Register to comment.