Permutation_cluster_1samp_test - inconsistent outcomes


Quick question: is there any good reason why I obtain different outcomes using permutation_cluster_1samp_test from MNE 0.19.2 to MNE 0.21?
More specifically, threshold is the same but the step-down-in-jumps iteration seems to work differently.
With MNE 0.19 there are no significant clusters while with the newer version there are 5 significant ones…
I am using exactly the same script with the two MNE versions (so same data, same parameters… same code!). …and Windows 10.

I did not find anything possibly related in the “What’s new” part of the documentation

This is kind of crucial information for me, I might need to re-run a bunch of analysis performed with the old version :confused:

My thanks in advance!


Hello, I don’t know the answer, but I’ll just tag @agramfort, @larsoner, and @drammock, who might know an answer. Please stay tuned :innocent:

I’ve realized that the two environments I was using have different numpy/scipy versions affecting data preparation upstream. Mystery solved then :relaxed: I apologize for bothering!

1 Like

No problem, thanks for following up @giuliagennari !