Software Library APIs: Lessons Learned from scikit-learn
summary
scikit-learn exposes a small set of concrete abstractions — transformers, estimators, meta-estimators — that map cleanly to ETL, fit/predict and validation; it uses semantic versioning for stability and composes via pipelines so a pipeline supports fit and predict like any estimator. civis ml extensions implements a stacking estimator that trains several base models (logistic regression gradient boosting random forest) on train/holdout folds and feeds their predictions as features to a meta-estimator and it works as a drop-in estimator.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:52What is an API
- 2:44How to build an API
- 6:09Building from the ground up
- 7:32What is a model
- 15:30Flexible and extendable
- 16:41Model stacking
- 21:38Conclusion
- 23:27Questions
related talks
33mUsing Scikit-Learn's interface for turning Spaghetti Data Science into Maintainable Software
20mTranslating from {tidymodels} and scikit-learn: Lessons from a 'bilingual' course
29mGetting Scikit-Learn To Run On Top Of Pandas
44mMachine Learning Made Easy With PyCaret
32mMachine Learning Infrastructure at Stripe: Bridging from Python JVM
01h 17mMachine Learning with Scikit-Learn
01h 24mInpsect and try to interpret your scikit-learn machine-learning models
37mSoftware Development for Machine Learning in Python