TickML is an open-source machine learning framework designed specifically for financial markets.
It provides a clean, modular, and extensible foundation for building, training, evaluating, and analyzing ML, Deep Learning, and Reinforcement Learning models on market data — without forcing rigid pipelines or assumptions.
The project focuses on architecture, realism, and research-grade workflows, rather than quick demos or toy examples.
- APIs may change
- Features are added incrementally
- Feedback and suggestions are welcome
- Stability will improve over time
This project is currently intended for:
- researchers
- practitioners
- advanced learners
- contributors interested in finance ML systems
TickML is built on a few core principles:
Data processing, feature engineering, normalization, modeling, evaluation, and backtesting are separate modules.
Users decide how to combine them.
There is no single “correct” workflow. Different strategies, models, and markets require different setups.
The framework is designed around:
- noisy data
- non-stationarity
- regime shifts
- realistic evaluation
- walk-forward thinking
Nothing is hidden. No silent preprocessing. No automatic magic. Every step is visible and configurable.
- Unified
BaseModelabstraction - Model registry for dynamic model loading
- Clear separation between ML / DL / RL models
- XGBoost classifier
- Hyperparameter optimization using Optuna
- Feature importance
- SHAP explainability
- JSON / console model summaries
- LSTM classifier (PyTorch)
- GPU / CPU support
- Optuna hyperparameter optimization
- Full model reports
- Confusion matrix, metrics, explainability hooks
- Confusion matrix (base-level, shared across models)
- Classification metrics
- Modular evaluation logic
- ❌ Not a one-click AutoML system
- ❌ Not a black-box trading bot
- ❌ Not a strategy marketplace
- ❌ Not optimized for retail trading hype
This framework is meant to be:
- transparent
- inspectable
- extendable
- research-friendly
- More deep learning models (CNN-LSTM, Attention)
- Unified reporting across all models
- Improved evaluation utilities
- Better examples & notebooks
- Reinforcement learning environments
- Walk-forward & regime-aware evaluation
- Backtesting integration
- Experiment tracking hooks
- Transformer-based models
- Multi-asset workflows
- Portfolio-level modeling
- Research reproducibility tools
Contributions are welcome, but architecture consistency matters.
If you plan to contribute:
- Open an issue first
- Explain the motivation clearly
- Keep changes modular
- Avoid adding hidden logic
This project values clean design over feature quantity.
At this stage, feedback is especially valuable on:
- API design
- architecture decisions
- evaluation philosophy
- missing core components
If something feels unclear or restrictive, that’s a signal — not a complaint.
This project is released under an open-source license. Details will be added as the project stabilizes.
TickML is not trying to replace existing libraries.
It exists to explore how finance ML systems should be structured when correctness, realism, and extensibility matter more than shortcuts.
If that resonates with you, you’re in the right place.
If you find this project useful, you can support it by:
- sponsoring development
- providing feedback
- reporting issues
- contributing code or documentation
Sponsorship helps keep the project maintained and evolving.