Planning_final_stock

Repo for final project at Planning Algorithms course at My University.

Description

We want to build an reinforcement learning algorithm to trade stocks with maximum efficiency. To do this, we use:

• Rapidly exploring random tree (RRT)
• Value Iteration (VI)

If we are good:

• Deep Q-learing
• Temporal difference
• Model-predictive control

Structure of our project

• data folder holds the data we train and test on
• utils contains some useful functions, i.e. to prepare data for models
• models folder has subfolders dedicated to each algorithm we use
• pipeline.ipynb is the main file which runs the whole stuff

Authors

Todo, see collaborators :D

For authors:

1. Our actions are: sell, buy, do nothing
2. Our environment is (vector of prices, vector of pct_changes, current day & day of week)
3. Observation is (10 last prices, 10 last pct_changes (9 mb?), current day of week, number of stocks on agent's balance, money he has)
4. Reward function: total cost (n_stocks*stock_price + curr_money) - start_money