Hands-On Machine Learning for Algorithmic Trading

RMSProp

RMSProp modifies AdaGrad to use an exponentially-weighted average of the cumulative gradient information. The goal is to put more emphasis on recent gradients. It also introduces a new hyperparameter that controls the length of the moving average.

RMSProp is a popular algorithm that often performs well, provided by the various libraries that we will introduce later and routinely use in practice.