Thank you for sharing your code to Nguyen 2020. In the script knockoffs/data_simulation.py, the variable "snr" does not corresponds to the signal-to-noise ratio as defined generally and the variable "snr" corresponds to the sqrt of the actual SNR.
This effects the x axis label in figure 2 in Nguyen 2020 which instead of 1,2,3,4,5 corresponds to 1,4,9,16,25. I noticed this problem when I observed I was able to get only ~20% power when snr is equals to 2 compared to ~70% reported in the paper.
(2020) Binh T. Nguyen, Jérôme-Alexis Chevalier, Bertrand Thirion and Sylvain Arlot. Aggregation of Multiple Knockoffs. Proceedings of the 37th International Conference on Machine Learning (ICML 2020), PMLR 119.