Writing

Build a Brawl Stars agent from recorded gameplay, exploring training-data collection, visual features, and supervised action prediction with neural networks. [Read More]

Learn policy gradients by building a trading agent, from the environment and reward design to policy-network training, practical challenges, and experimental results. [Read More]

Build the same price-prediction neural network with NumPy, TensorFlow, and Keras to see what each abstraction handles and how training works underneath. [Read More]

The A3C method in Reinforcement Learning (RL) combines both a critic’s value function (how good a state is) and an actor’s policy (a set of action probability for a given state). I promise this explanation doesn’t not contain greek letters or calculus. It only contains English alphabets and subtraction in... [Read More]

I’ve been bugged by the native state management system in React that I finally had to take a stab at Redux. Here are some notes I took along the way to understand what Redux is and why we need it. [Read More]