- make machine
- learning models
- deployment and monitoring
- easiest for everyone
Eurybia is a Python library dedicated to the monitoring of Data Science models. It provides several types of visualizations that display through an HTML report (or directly in notebook mode) which help in detecting drift (data drift & model drift). It also support data validation before putting a model into production.
Features
- Consistency analysis between the baseline dataset and the current dataset
- Performance of the data drift classifier
- Feature importance: features that discriminate the most two datasets
- Scatter plot: Putting the drift of a variable into perspective with its importance in the deployed model
- Dataset analysis: distribution comparison between variable from the baseline dataset and the current dataset
- Predicted values analysis: distribution comparison between predicted probabilities from the baseline dataset and the current dataset
- Features contribution for the data drift classifier
- AUC evolution: performance comparison for the data drift classifier over time
- Offers several parameters in order to summarize drift in the most suitable way for your use case
- Model performance evolution: compare your model performances over time
