A quick attempt to trade with Reinforcement Learning
Using Q-learning for training with a personalized reward scheme. See: https://keras.io/examples/rl/deep_q_network_breakout/

• learn.py: used to build a model from provided data.
• evaluate.py: trade on the provided data and return a plot.
• find_best.py: trade on the provided data using all trained models and find the best one.

• t-bot_112.h5: trained on 14 days of ADA data using only the bet_USD column. It achieves 12% gain with 50% sampled selling and 9% without sampling.
• t-bot_163.h5: same performance but perfrms better on a daily basis.

The provided data is sampled with a 1 minute resolution.

The various constants set in the learn script can be cahnged and are based on experimenting with the data. Currently, the used ones are the best performing.