Welcome to Gradient Trader - a cryptocurrency trading platform using deep learning. We are four UC Berkeley students completing our Masters of Information and Data Science. Some of us come from a finance background, others with expertise in deep learning / reinforcement learning, and some are just interested in the cryptocurrency market. With that, we’d like to present our work to you and hope you’ll share with others that may find this blog of interest!

Reinforcement learning (RL) is a type of machine learning that allows the agent to learn from its environment based on a reward feedback system. One of the most well known examples of RI is AlphaGo, developed by Alphabet Inc.’s Google Deepmind. It was trained using a number of machine learning models, including RI, to learn how to play the notoriously challenging board game Go and went on to beat the world’s greatest players. This feat was thought of as impossible with the technology available at the time.

Bitcoin, Ethereum, and Ripple have been selected as representatives in our EDA. Bitcoin and Ethereum well represent mainstream cryptocurrencies that most people are trading with, while Ripple represents new emerging cryptocurrencies with high potentials.

How do we create a programmatic environment to explore the Cryptocurrenct environment?

We need to keep track of the performances of our trader. And more importantly, understand how to reward our trader for making good trade decisions.

So how exactly does an agent trade? Here we explain our DQN agent.

And finally, how are we evaluating our results?