Wrong squaring in ndPAC? about tensorpac HOT 3 CLOSED

@raphaelvallat indeed results are different.

The stats works because it's following what's described in the paper. However, the pac estimation returned by the function doesn't corresponds to the one described in the paper. The s is squared in Ozkurt formula, but the returned pac should not. Actually, if you take your values, square it and multiply by the number of time-points (4000) you should retrieve my values.

So yes, please make a PR !

Merci,

++

from tensorpac.

I see your point. I don't really remember why I didn't strictly follow the paper but at the ed I think it's going to produce the same results because :

s > 2 * xlim
<=> s > 2 * N * erfinv(r) ²
<=> s / N > 2 * erfinv(r) ²

That's why the xlim doesn't contains N. An advantage of the current implementation is that I don't need to make a copy of pac (your s = pac ** 2).

However, I might also miss something. Could you please try it on your side with your implementation to double check?

from tensorpac.

Hi @EtienneCmb ! So I checked and it does induce produce different results. It seems that the current implementation gives PAC values that are not bounded by [0-1]. The masking seems correct however.

Let me know what you think of the notebooks below, and if that sounds good I can submit a PR

Notebook = ndpac.pdf

from tensorpac.