Cycles in the final prediction

[SPChytas]

Hello, I have tried to run the nonlinear version of NoTears for a lot of different datasets, and when I am using a w_threshold < 0.3 (e.g 0.1, 0.2) the final predicted graph has, in many cases, cycles in it, no matter the acyclicity parameters values (h_tol, rho_max). Actually, there are cases in which even w_threshold = 0.3 results also to cycles.