The Data and Network Introspection Kit (DNIKit) has officially been open sourced! DNIKit is a Python toolkit for analyzing machine learning models and datasets, and can be used to:
- create a comprehensive dataset analysis report
- identify duplicate data samples
- find rare data samples, annotation errors, or model biases
- compress networks
- detect inactive units in a model (e.g., due to dying ReLU)
- …and more.
To learn more and get started, check out the DNIKit Documentation.