stockly

A complete stock market prediction and backtesting system with SQL-based data storage, multiple ML models (baseline + LSTM), and a retro pixel-style visualizer.

Live Demo

Try the interactive Streamlit app: https://sstockly.streamlit.app/

Technologies

• Python - Primary language
• pandas - Data manipulation
• numpy - Numerical computing
• scikit-learn - Baseline models (Logistic Regression, Random Forest)
• TensorFlow/Keras - LSTM/GRU deep learning models
• SQLite - Relational database for data storage
• matplotlib - Visualization with retro pixel theme
• Streamlit - Interactive demo app

Project Structure

stock-predictor/
├── data/
│   ├── raw/              # Raw CSV files
│   ├── market.db         # SQLite database
│   └── processed/        # Exported datasets
├── notebooks/
│   ├── 01_exploration.ipynb
│   └── 02_model_prototyping.ipynb
├── src/
│   ├── config.py
│   ├── database/
│   │   ├── schema.sql
│   │   └── db_utils.py
│   ├── data_acquisition/
│   │   ├── fetch_prices_api.py
│   │   └── load_prices_csv.py
│   ├── data_preprocessing/
│   │   ├── clean_prices.py
│   │   ├── calculate_technical_features.py
│   │   └── create_targets.py
│   ├── models/
│   │   ├── build_datasets.py
│   │   ├── sequence_dataset.py
│   │   ├── train_baseline_models.py
│   │   ├── train_lstm.py
│   │   └── time_series_backtest.py
│   ├── visualization/
│   │   ├── style_pixel_theme.py
│   │   ├── plot_price_and_signals.py
│   │   └── plot_performance.py
│   └── app/
│       └── streamlit_app.py
├── models/               # Saved trained models
├── tests/
└── requirements.txt

Database Schema

The SQLite database (data/market.db) contains:

• symbols - Stock ticker information
• prices - Historical OHLCV data
• features - Engineered technical indicators
• targets - Next-day returns and direction labels
• predictions - Model predictions

Setup & Installation

1. Clone Repository

git clone https://github.com/AliAbouelazm/stock-predictor.git
cd stock-predictor

2. Create Virtual Environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

Data Acquisition

Option 1: API (Alpha Vantage)

1. Get API key from Alpha Vantage
2. Set environment variable:

   export ALPHA_VANTAGE_API_KEY="your_key_here"

3. Fetch data:

   python src/data_acquisition/fetch_prices_api.py

Option 2: CSV Files

Place CSV files in data/raw/ with columns: date, open, high, low, close, adjusted_close, volume

Then load:

python src/data_acquisition/load_prices_csv.py

Option 3: Sample Data

Create sample data for testing:

python create_sample_data.py

Pipeline Steps

1. Initialize Database

python -c "from src.database.db_utils import get_connection, initialize_schema; conn = get_connection(); initialize_schema(conn); conn.close()"

2. Load Prices

Use one of the data acquisition methods above.

3. Compute Features

python src/data_preprocessing/calculate_technical_features.py

4. Create Targets

python src/data_preprocessing/create_targets.py

5. Train Models

Baseline Models:

python src/models/train_baseline_models.py

LSTM Model:

python src/models/train_lstm.py

6. Run Streamlit App

streamlit run src/app/streamlit_app.py

Models

Baseline Models

• Logistic Regression - Multinomial classification for direction prediction
• Random Forest - Ensemble classifier

Deep Learning

• LSTM - Long Short-Term Memory network for sequence-based prediction
  • Uses 30-day lookback window
  • 2-layer LSTM architecture
  • Predicts direction (-1, 0, 1)

Features

Technical indicators computed:

• Daily returns (1d, 5d)
• Rolling volatility (10d, 20d)
• Simple Moving Averages (10, 20, 50 days)
• RSI (14-period)
• MACD and signal line
• Lagged returns

Backtesting

The system includes backtesting functionality:

• Time-based train/test splits
• Strategy performance vs buy-and-hold
• Metrics: Total return, Sharpe ratio, Max drawdown

Visual Style

Charts use a retro pixel aesthetic:

• Blocky square markers
• Bold color palette (green/red/yellow/blue on black)
• Thick gridlines
• Step-style plots
• Monospace fonts

Limitations

• No transaction costs modeled
• Overfitting risk with limited data
• No fundamental data or news sentiment
• Educational purposes only - not financial advice

License

This project is for educational purposes.

Author

AliAbouelazm