We wanted to make it easier for people to get into crypto.
What it does
It is a reinforcement bot that trades cryptocurrencies. We started with bitcoin but it can be applied to any of the altcoins as well.
How we built it
The reinforcement bot is made using the Tensortrade library which allows for modular agents that can be placed in many environments. We used a DQN agent built using TensorFlow. The observations consisted of a matrix of 5 columns which were the difference with lag 1, MACD, RSI14, and 2 moving averages(10 and 50).
Challenges we ran into
We couldn't figure out how to actually deploy the bot but we figured it out by treating the agent as any other TensorFlow model.
Accomplishments that we're proud of
We are proud that we got the bot to actually work within the environment. As well, it was able to perform pretty well.
What we learned
We learning a lot about reinforcement learning. From making the environments, the agents, and tuning the model as well. We also learned about deploying a reinforcement learning agent to take actions based on observation.
What's next for TradeSmart
We hope to do a lot more tests to see if it's actually viable then maybe even let it do some real trading!