Simulation Environment for stock market and agent which takes the input as Markov decision process. The input is mathematical indicators deriven from price and volume of stock which passes through a neural network that tries to optimize alpha generated and minimize risk.
- Simulation environment
- Agent interaction
- W&B data and model versioning
- Streamlit for analysis for training - 25%
- Dynamic Clustering of stocks
- Parallel environment
- UI for agent training
- Pretrain from date to date
- Pretrain with Expert dataset
- Multi input dictionary from env
- Dictionary observation for prices / indicators
- Time window input
- Train process with time window input
- Take flag days out of environment
- Cirriculum learning
- Learn not sell for loss for generalization
- Learn always to keep 50.000$
- Validation of model for different time horizons for PBT
- Day Trading Environment
- MARL
- Ensemble voting for trading strategy
- Kelly Criterion Auxilary task
- Data Selection Add to Preprocess
- Correlation Table as input
- Past prices for the input
- Avg. bought price of stocks in environment as a fuction
- Dynamic Clustering
- Normalized inputs for the model
- Switching stocks and keeping the same stock holding if is in same list
- pip install -r requirements.txt
- StockTradingBot.ipynb notebook is for training process