[ANSWERED]: use local random seed in XValidation operator

[jaysonpryde]

Hi,

     I will just ask what's the "use local random seed" and "local random seed" parameters of XValidation (and other operators) for?
     Specifically, what's their effect on the result of validation? Are they just used for the randomness of the data splitting/sampling
     in validation?

Thank you!

[Marius]

Hi,
 
RapidMiner by default uses a pseudo random number generator (rng) to provide random numbers to any operators which need random numbers. The sequence of random numbers is identical for each process run, if and only if the so-called seed of the RNG is the same.
By default, RapidMiner uses a fixed seed per process, that means, that if you run a process twice you will get the same results, even if your operators use randomness (like the X-Validation does for splitting).

You can override the process-global seed per operator, and that is done by the parameters "use local random seed" and "local random seed". Thus, if you change the local seed of the X-Validation, it chooses different splitting points, and the performance values will differ (usually not much, though). Usually, there is not need to touch the local seed parameter; otherwise you run into the danger of "optimizing" and "tweaking" your process by using different numbers.

Best, Marius

[jaysonpryde]

Thanks a lot for the reply Marius