Hyperparameter Tuning With Policy Gradients From Scratch

We derive the policy gradients algorithm, implement it from scratch in Python, and use it to tune hyperparameters for machine learning models. Machine learning algorithms contain two different types of parameters: those that are computed during model training, and those that aren't (hyperparameters). It can be difficult to find good hyperparameters for a particular algorithm, and their values often greatly affect the resulting model's performance. In this article, we'll explore the policy gradients algorithm for meta learning of hyperparameters. The policy gradients algorithm was created for solving reinforcement learning (game playing AI) problems.