🚀 Try the live dashboard: Market Predictions Dashboard

Test Predictions Dashboard showing model forecasts and actual market data

📺 Watch Workshop Recording Coming Soon

This repository contains materials for the Machine Learning Workshop focused on financial time series prediction. The workshop explores three different approaches to predicting stock prices: ARIMA, Prophet, and Deep Neural Networks.

Hi I'm Ivan Brigida an Analyst at Google with a background in Economics and Data Analysis. Before joining Google I worked in the banking sector and 3 years ago I started investing in U.S. markets while living in Ireland. My work and investing experience focus on using economic data to understand the economy, make financial decisions, and develop market strategies. You can read my latest article about using LLMs for long-term financial news analysis, and find more articles at pythoninvest.com/blog.

Last year I ran the Stock Markets Analysis Zoomcamp which attracted 1800+ registrations. I'll be running it again in 2025 you can express your interest here:

Help us grow and improve PythonInvest:

1. Engage: 🌟 Star this repo, 👍 like the video and 💬 comment on YouTube.
2. Contribute: 🍴 Fork the repo, try it, tweak the indicators and share feedback with a link.
3. Donate: ☕ Buy Me a Coffee or ❤️ GitHub Sponsorship.

This workshop builds on our previous workshop which focused on data collection and visualization. The Economics Dashboard from that workshop is still updated weekly through GitHub Actions and runs on Streamlit Cloud, demonstrating the full pipeline from data collection to deployment. Similarly, this workshop's Market Predictions Dashboard is also updated daily with the latest predictions.

Like the previous workshop, this project extensively utilized AI-assisted development (~$112 in AI costs) using:

• Cline Extension for VS Code
• Anthropic Claude 3 Sonnet (20241022)

The AI assistance accelerated:

• Code implementation and replication across models
• Documentation generation and maintenance
• Model comparison and analysis
• Bug fixing and optimization

For details on how AI was used in development, see REPRODUCTION.md.

While the previous workshop covered Docker containerization and cloud deployment options, this session focuses purely on modeling techniques. We'll use Streamlit Community Cloud for simple deployment, without diving into Docker or other cloud platforms. This allows us to concentrate on the core machine learning concepts and model comparison.

• Compare traditional and modern approaches to time series prediction
• Understand the strengths and limitations of each model type
• Learn practical feature engineering for financial data
• Develop intuition for model selection and evaluation

The workshop demonstrates a progression from simple to complex modeling approaches:

• Quick-start solution with minimal code
• Built-in trend and seasonality decomposition
• Simple parameter tuning and configuration
• Excellent baseline for time series prediction

• Classic ARIMA(2,1,2) approach
• Strong theoretical foundation
• Detailed model interpretation
• Integration with market indicators

• Sophisticated feature engineering
• Careful architecture optimization (8->8 units)
• Returns-based prediction approach
• Complex regularization and training process

• Trading metrics implementation
• Risk-adjusted returns analysis
• Performance comparison methodology

For detailed implementation steps and code examples, see REPRODUCTION.md.

• Python 3.12 or higher
• TensorFlow 2.16.2 (stable version with Python 3.12 support)
• Additional packages listed in requirements.txt

1. Create a virtual environment:

python -m venv .venv

2. Activate the virtual environment:

• On Windows:

.venv\Scripts\activate

• On macOS/Linux:

source .venv/bin/activate

3. Install the required packages:

pip install -r requirements.txt

.
├── app/
│   └── main.py                 # Streamlit dashboard application
├── data/
│   └── market_data.db         # SQLite database with market data and predictions
├── notebooks/
│   ├── prophet_experiment.ipynb  # Quick-start Prophet analysis
│   ├── arima_experiment.ipynb    # Statistical ARIMA modeling
│   └── dnn_experiment.ipynb      # Advanced Deep Learning approach
├── scripts/
│   ├── init_db.py             # Database initialization script
│   ├── update_data.py         # Market data update script
│   ├── train_models.py        # Model training pipeline
│   └── models/
│       ├── arima_model.py     # ARIMA model implementation
│       ├── prophet_model.py   # Prophet model implementation
│       └── dnn_model.py       # Deep Neural Network model
├── static/
│   └── images/                # Project images and screenshots
└── .github/
    └── workflows/             # GitHub Actions automation

After installing the requirements, you can:

1. Initialize the database:

python scripts/init_db.py

2. Update market data:

python scripts/update_data.py

3. Train models and generate predictions:

python scripts/train_models.py

4. Run the Streamlit dashboard:

streamlit run app/main.py

The dashboard will be available at http://localhost:8501

• Win Rate: ~51.2%
• Market outperformance: +1.1%
• Average returns: 0.21%
• P/L ratio: 1.15

• Win Rate: ~54.5%
• Market outperformance: +3.3%
• Average returns: 0.42%
• P/L ratio: 1.45

• Win Rate: ~58.1%
• Market outperformance: +1.3%
• Average returns: 0.38%
• P/L ratio: 1.12

We welcome contributions! Here's how you can help:

• Star ⭐ this repository to show your support
• Submit issues for bugs or enhancement ideas
• Fork the repository and submit pull requests
• Share your model improvements and feature ideas

See the LICENSE file for details.