zkML-Noir

Description

Python ML model transcoding Noir, including various algorithms such as Decision tree, K-Means, XGBoost, FNN, CNN

Directory structure

• data: ML model training and prediction data, including raw data and pre-processed data
• model: Three types of ML models
• noir: zkML generated Noir code
• tests: Testcase for zkML transpiler code
• zkml: The zkML code generation and floating point numbers quantize integer numbers
  • decision_tree: Python2Noir generate Noir prediction code for the decision tree based on sk-learn library
  • k_Means: Python2Noir generate Noir prediction code for the center points based on sk-learn library
  • quantization: ML floating point numbers quantize integer numbers
  • routine_code_generate: Routine generate Noir prediction code for the CNN and RNN based on Pytorch library
  • XGBoost: Python2Noir generate Noir prediction code for the XGBoost classification and regression based on XGBoost library

Build guide

• Python 3.7+
• Anaconda

Import package

• python2noir
• joblib
• scikit-learn
• xgboost
• numpy
• unittest
• pandas
• pytorch
• torchvision

Usage

git clone https://github.com/storswiftlabs/zkml-noir.git
cd zkml-noir
# execute decision tree generate code
python  -m unittest tests/zkml/decision_tree/test_decision_tree_to_noir.py

# execute K-Means generate code
python  -m unittest tests/zkml/k_Means/test_k_Means_to_noir.py

# execute XGBoost generate code
python  -m unittest tests/zkml/XGBoost/test_xgboost_to_noir.py

# Train the CNN model
python tests/zkml/cnn/mnist_cnn.py
# execute FNN generate code
python zkml/routine_code_generate/fnn_to_noir.py
# execute CNN generate code
python zkml/routine_code_generate/cnn_to_noir.py
# Load the model and extract inputs
python zkml/routine_code_generate/extract_inputs.py