Watch this video on YouTube
Why might the loss not decrease monotonically with each run when using stochastic search algorithms like those in Hyperopt ?