PPO Implementation in PyTorch

In this blog post, we will explore the Proximal Policy Optimization (PPO) algorithm. We’ll compare it to other deep reinforcement learning algorithms like Double Deep Q-learning and TRPO. Additionally, we’ll learn how to implement PPO using PyTorch.

Double Deep Q-Network

In double DQNs, we use a separate network to estimate the target rather than the prediction network. The separate network has the same structure as the prediction network. And its weights are fixed for every T episode (T is a hyperparameter we can tune), which means they are only updated after every T episode. The update is simply done by […]

Climbing the Mountain with Neural Network

Function Approximation For problems with very large number of states it will not be feasible for our agent to use table to record the value of all the action for each state and make its policy accordingly. In Function approximation agent learns a function which will approxmately give it best action for particular state. In this example we will use […]

SARSA in the Wind

We will use SARSA algorithm to find the optimal policy so that our agent can navigate in windy world. SARSA State–action–reward–state–action (SARSA) is an algorithm for learning a Markov decision process policy, used in the reinforcement learning area of machine learning. SARSA focuses on state-action values. It updates the Q-function based on the following equation: Q(s,a) = Q(s,a) + α (r + γ Q(s’,a’) – Q(s,a)) Here s’ […]