Seed and repricability

[Thomas Feutrier]

Hello, I have a little problem, when I run irace with a seed set twice, I get different results.
My solver, if irace gives it the same seed twice in a row, returns exactly the same result, so the problem doesn't seem to come from there.
I looked at the instances used, and irace samples exactly the same ones.
I'm using an older version of irace which doesn't remember the results of each run in the logfile (cluster version).
For the first iteration we have exactly the same results but from the second it starts to vary slightly until we have very different configurations.
I wanted to know if the generation of random configurations with the models was controlled by the seeds? And if you have any idea where this comes from?

[Manuel López-Ibáñez]

Hi Thomas,

irace uses the random number generator of R and it should be able to reproduce exactly the same output if the target-runner does the same. Yes, this also applies to the generation of random configurations 

If you could share the .Rdata files of the two runs that differ, I may be able to tell you more.

Cheers,

Manuel.

[Manuel López-Ibáñez]

Thomas shared his .Rdata files and it seems his target-runner is not returning the same output for the same configuration, instance and seed. It is not a bug in irace.