FKLearn, A functional machine learning library
summary
A practical demo of FKLearn that claims to make models pure functions so they are models as functions and easily serialized for production; it actually wraps scikit‑learn and LightGBM (and mentions xgboost) and shows a credit‑card spend predictor built with pandas, time/space splits, extensive validators and learning‑curve experiments, backed by Nubank usage and >1,000 GitHub stars. Code produces a stateless predict function plus logs for feature importance and seems to work end‑to‑end.
AI-written summary · about
chapters (11)
- 0:00FKLearn introduction
- 3:21Why FKLearn at Nubank
- 8:50Dataset for tutorial
- 12:29Splitting data space-time
- 16:03Functional pipeline in FKLearn
- 22:14Training prediction functions
- 25:27Validating models with FKLearn
- 28:39Complex validations beyond metrics
- 34:21Spatial learning curves
- 40:07Advanced validations and features
- 42:59Q&A
related talks
01h 02mMachine Learning with F#, Redux
21mWhat, Why and How of Federated Machine Learning – Implementation Using...
36mBuild prod ML systems with feature /training /inference pipelines
22mReal-world applications of machine learning: Beyond the hype
01h 12mBuild Simpler Production ML Systems using Feature/Training/Inference Pipelines
28mEnhancing Machine Learning Workflows with skore
21mKeynote: Easier Machine Learning Thoughts From Scikit-Learn
35mScikit-Learn can do THAT?!