To reproduce the main experiments, run reproduce_main_experiments.py. The seed is fixed, so you should be able to reproduce the results exactly.

To reproduce the figures from the paper with the slurm logs from the main experiments, run SelectGlobalMinBasedOnValues.ipynb.

Replace variable new_function with your function of interest to get results for a given function. 

To reproduce the baseline of training the neural network on randomly sampled points only, set the variable threshold_epoch to be equal to a large number (e.g. 4000). This variable indicates the epoch when the neural network will train on the trajectories from the previous epoch as opposed to randomly sampled points.




To reproduce the baselines with the Adam optimizer, just run what_adam_cant_do_thorough.py





To reproduce the experiments on which benchmark functions are "learnable" by a fully connected neural network, run testing_funcs_on_fcnn.ipynb.
