Hi, I am wandering if the randonization algorithm in classification is changed since June this year? I have results of many repetition basing on the same training fields (classes are not homogeneous (there are crops farmers declaration)).
Classifications were performed in June and in November, now I have SNAP 8.0 in June it was updated but I do not remember version.
I have ca. ca 3000 agriculture parcels = 1.4 mln pixels, always the same.
I perform classification basing on default 5000 (I have tested more ofcourse) but in June the repetitions of maxlike classification deliver very simmilar results, RF of course not. But now the maxlike results differ significantly.
I checked the code of the RF and MaxLikelihood classifiers and no changes in the randomization algorithm has been made. However I will look more in details about any possible changes in the engines that could have lead to changes in the random seed initialization. Another issue could be in the operator execution, if I am not wrong it is executed by product tiles in a parallel asynchronous mode with no guarantee of order execution (so each time it can be different).
I will share my findings as soon as possible.
Thanks for your answer. I test the multitemporal, optical and radar merged images classifications for crops recognition in control system IACS, its repeatability and accuracy assessment. I am preparing paper and don’t want to draw wrong conclusions