Inquiry about risk curve

[pipilurj]

Dear author,
Thank you very much for the code, I have a problem when running risk_curve.py, the results I got after running the code using default setting didn't have the double descent behavior, and the eval loss is smaller than your given results, I wonder if there is any difference between our hyper-parameter settings. e.g sample_size = 4000, epochs = 500, lr=0.001, batchsize=128
Thank you very much.