AI Trading Bot

Python project for Nifty options research: collect Dhan option-chain data, train a PPO reinforcement-learning model, then optionally paper or live-deploy it.

Generated data, models, and logs are not in git. A fresh clone has scripts and config only.

What is in this repo

ai_trading_bot/
├── .env.example              # Copy to .env; never commit .env
├── .gitignore
├── LICENSE
├── README.md
├── requirements.txt
├── setup.py
├── test_imports.py
├── run_script.sh             # Server cron helper (2025 holidays, conda paths)
├── ai_trading_bot_logrotate
├── processed_data/.gitkeep   # CSVs are created at runtime
└── scripts/
    ├── data_collection.py    # Dhan option chain + PCR metrics
    ├── data_processor.py     # Archive / trim training CSVs
    ├── trading_env.py        # Gymnasium env (hold / buy / sell)
    ├── train_model.py        # PPO training
    ├── hyperparameter_tuning.py  # Optuna
    ├── validate_model.py     # Standard + time-series validation
    ├── deploy_model.py       # Live/paper deploy
    ├── analyze_trades.py     # Reads logs/paper_trades.csv
    ├── visualization.py      # Training plots from models/training_metadata.json
    └── utils.py

These directories are created when you run the scripts; they are gitignored:

• processed_data/ — training_data.csv, processed_data.csv, archive/
• pre_processed_data/ — raw collector output
• models/ — PPO zip, VecNormalize, param JSON
• logs/ — train/eval logs, deployment trade history
• optuna/ — study databases and plots

Configuration

cp .env.example .env

Variable	Required for	Notes
DHAN_CLIENT_ID	data collection, deploy	Dhan client id
DHAN_ACCESS_TOKEN	data collection, deploy	Dhan API token
DHAN_ACCESS_TOKEN_EXPIRY	data collection	YYYY-MM-DD; used for expiry warnings
TELEGRAM_BOT_TOKEN	optional	Alerts
TELEGRAM_CHAT_ID	optional	Alerts

Never commit .env.

Install

TA-Lib is required by requirements.txt.

macOS:

brew install ta-lib

Then:

git clone https://github.com/KushalAzza/ai_trading_bot.git
cd ai_trading_bot
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
pip install -e .
cp .env.example .env

setup.py installs a shorter dependency list than requirements.txt. Use requirements.txt for a full run (Dhan, Optuna, torch, plotly).

Usage

Run in this order. trading_env.py is imported by train, validate, and deploy; it is not a CLI. test_imports.py is an optional smoke test.

flowchart TD
    env["Copy .env.example to .env"] --> collect["1. python scripts/data_collection.py"]
    collect --> processed["Writes processed_data/processed_data.csv"]
    processed --> archiver["2. python scripts/data_processor.py"]
    archiver --> training["Writes processed_data/training_data.csv"]
    training --> tune{"Tune hyperparameters?"}
    tune -->|Optional| optuna["python scripts/hyperparameter_tuning.py"]
    tune -->|Skip or after Optuna| train["3. python scripts/train_model.py"]
    optuna --> train
    train --> model["Writes models/ including training_metadata.json"]
    model --> validate["4. python scripts/validate_model.py"]
    model --> viz["Optional: python scripts/visualization.py"]
    model --> deploy["5. python scripts/deploy_model.py"]
    deploy --> trades["Writes logs/paper_trades.csv"]
    trades --> analyze["6. python scripts/analyze_trades.py"]

Loading

Cron: run_script.sh only runs data_collection.py during market hours. Set PROJECT_ROOT if the project is not at $HOME/ai_trading_bot. Holiday dates in that script are for 2025.

python scripts/data_collection.py
python scripts/data_processor.py

python scripts/hyperparameter_tuning.py --trials 5 --jobs 1
python scripts/hyperparameter_tuning.py --timeout 14400 --jobs 4 --train

python scripts/train_model.py
python scripts/train_model.py --optimize
python scripts/train_model.py --retrain

python scripts/validate_model.py
python scripts/visualization.py
python scripts/deploy_model.py
python scripts/analyze_trades.py
python test_imports.py

License

MIT. See LICENSE.