QuantIQ is a unified Python library for building robust, uncertainty-aware deep learning systems. It brings together lightweight, modular tools to help researchers and practitioners gain insight into model reliability and risk through principled uncertainty estimation.
Quantiq currently includes two powerful uncertainty quantification tools:
A plug-and-play PyTorch/HuggingFace wrapper for Monte Carlo Dropout–based uncertainty estimation in Transformers.
- Supports classification, regression, QA, and token tagging
- Computes entropy, confidence, and class-wise variances
- Enables dropout during inference for Bayesian-style sampling
form quantiq import DropWiseA deep ensemble wrapper for PyTorch models with support for adversarial training and dual-mode (epistemic + aleatoric) uncertainty estimation.
- Works with any PyTorch model
- Enables risk scoring and calibration
- Includes built-in visualization and prediction APIs
form quantiq import SmartEnsemble📖 Full SmartEnsemble Documentation
pip install quantiqOr install from source:
git clone https://github.com/aryanator/QuantIQ.git
cd quantiq
pip install -e .- Safety-critical predictions (medical AI, self-driving, finance)
- Uncertainty-aware active learning
- Robust ML pipelines with explainable confidence
- Research experiments involving confidence, entropy, risk
Explore examples, API usage, and task-specific walkthroughs in the GitHub repository:
🔗 https://github.com/aryanator/QuantIQ
MIT License
Built by Aryan Patil to make uncertainty estimation simpler, smarter, and production-ready.