Package to calibrate and understand ML Models
Welcome to ML-Insights!
This package contains two core sets of functions:
For probability calibration, the main class is SplineCalib. Given a set of model outputs and the “true” classes, you can fit a SplineCalib object. That object can then be used to calibrate future model predictions post-hoc.
>>> model.fit(X_train, y_train) >>> sc = mli.SplineCalib() >>> sc.fit(X_valid, y_valid) >>> uncalib_preds = model.predict_proba(X_test) >>> calib_preds = sc.calibrate(uncalib_preds)
>>> cv_preds = mli.cv_predictions(model, X_train, y_train) >>> model.fit(X_train, y_train) >>> sc = mli.SplineCalib() >>> sc.fit(cv_preds, y_train) >>> uncalib_preds = model.predict_proba(X_test) >>> calib_preds = sc.calibrate(uncalib_preds)
For model interpretability, we provide the ice_plot and histogram_pair functions as well as other tools.
>>> rd = mli.get_range_dict(X_train) >>> mli.ice_plot(model, X_test.sample(3), X_train.columns, rd)
>>> mli.histogram_pair(df.outcome, df.feature, bins=np.linspace(0,100,11))
Please see the documentation and examples at the links below.
We have tested this tool to the best of our ability, but understand that it may have bugs. It was most recently developed on Python 3.7.3. Use at your own risk, but feel free to report any bugs to our github. <https://github.com/numeristical/introspective>
$ pip install ml_insights
>>> import ml_insights as mli >>> xray = mli.ModelXRay(model, data)
>>> rfm = RandomForestClassifier(n_estimators = 500, class_weight='balanced_subsample') >>> rfm_cv = mli.SplineCalibratedClassifierCV(rfm) >>> rfm_cv.fit(X_train,y_train) >>> test_res_calib_cv = rfm_cv.predict_proba(X_test)[:,1] >>> log_loss(y_test,test_res_calib_cv)
Find the latest version on github: https://github.com/numeristical/introspective
Feel free to fork and contribute!
Free software: MIT license
Alex Goldstein, Adam Kapelner, Justin Bleich, and Emil Pitkin. 2014. Peeking Inside the Black Box: Visualizing Statistical Learning With Plots of Individual Conditional Expectation. Journal of Computational and Graphical Statistics (March 2014)
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Hashes for ml_insights-1.0.1-py2.py3-none-any.whl