Reproducing benchmark results

[aleflabo]

Hi all,

I'm currently trying to reproduce the paper results for the MSL dataset following the code and the hyperparameters present on this issue by @KSGulin and with the correction proposed by @sarahmish. At the moment I've a final score of 0.5747 instead of the paper one 0.623, which means that there is a drop in performances. Could this be addressed to the training randomness or there is something else to change to have the same results as in the paper?

Originally posted by @aleflabo in https://github.com/signals-dev/Orion/issues/221#issuecomment-845233097