ModelTrace Documentation

Research-grade ML Model Inspector, Debugger & Robustness Analyzer

Welcome to ModelTrace

ModelTrace is a research-oriented framework for analyzing, debugging, and evaluating machine learning models.

Use the links above to explore:

Quick Start

git clone https://github.com/MANISH-K-07/ModelTrace.git
cd ModelTrace
pip install -r requirements.txt
python setup.py install
# Run all experiments
python examples/run_all_experiments.py

Example Output


============================================
     Running ALL ModelTrace Experiments
============================================

[Experiment 1] Sparsity Analysis
Conv sparsity (avg): 0.0000
FC sparsity (avg):   0.0000
--------------------------------------------

[Experiment 2] Activation Statistics
conv1: mean=-0.0926, std=0.6074, max=1.5994
conv2: mean=0.0355, std=0.2371, max=0.8379
fc: mean=0.0229, std=0.1267, max=0.1856
--------------------------------------------

[Experiment 3] Distribution Shift Test
Normal mean activation magnitude:  0.2797
Shifted mean activation magnitude: 0.3642
--------------------------------------------

[Experiment 4] Activation Drift
Drift score: 0.1093
--------------------------------------------

[Experiment 5] Model Regression Test
Regression drift score: 0.3458
--------------------------------------------

[Experiment 6] Layer Sensitivity Analysis
conv1: sensitivity=0.1984
conv2: sensitivity=0.0725
fc: sensitivity=0.0292
--------------------------------------------

[Experiment 7] Failure Attribution
Output shift magnitude: 0.0257
Layer-wise contribution:
conv1: 0.6609
conv2: 0.2325
fc: 0.1066
--------------------------------------------

[Experiment 8] Model Robustness Score
FINAL ROBUSTNESS SCORE: 0.0646
--------------------------------------------

✅ All experiments completed successfully.
    

Author

Manish Krishna Kandrakota – Final year B.Tech CSE student at Sreenidhi Institute of Science and Technology, Hyderabad.
GitHub: https://github.com/MANISH-K-07