Hi all,
I think this might be interesting to some of you: I added rule creator dashboard to my rule-estimator library.
This is a dash app that allows you to interactively create business rules (basically if-then statements) to generate predictions for a given dataset.
This is what people used to do before machine learning became popular. However, business rules have the advantage that they are often much easier to understand and communicate. Creating good business rules can be a challenge and that’s where my dashboard comes in.
You can:
- work interactively with your data to create new business rules
- add rules based on a single feature or multiple features at once using the parallel plot
- get suggestions for good rules to try
- inspect performance of the overall model and individual rules
- export the rules to a scikit-learn compatible estimator
- export to pickle or to a transparent and editable .yaml file
Inspired by Vincent Warmerdam’s CSVconf talk and human-learn library, this is basically a dash implementation of the ideas outlined there.
You can generate a dashboard for you own dataset using the rule-estimator library: GitHub - oegedijk/rule_estimator: A scikit-learn compatible estimator based on business-rules with interactive dashboard included
Deployed example can be found at: https://ruledashboarddemo.herokuapp.com
Vincent’s csvconf talk: Vincent Warmerdam - Human-Learn: Let's Draw Machine Learning Models and Train Humans Instead - YouTube
