Reproducing the same hyper-parameter grid search results with every run

Hi everyone,
I am looking to perform grid search for all the hyper-parameters in any BNN method such MC-Dropout (drop-out rate, learning rate, weight decay for L2, no. of stochastic runs etc,…). I have fixed the random seed for e.g. (rng = np.random.RandomState(2)), but I am not able to reproduce the same results. Can anybody help me on how to reproduce the same results every-time I run my python code (uses python modules).

The Reproducibility docs give you some information on how to get deterministic results.

1 Like