https://csvfit.com/blog Train an ML model on your CSV — sklearn and other ML packages running entirely in your browser via Pyodide + WebAssembly. No server, no uploads. en-us https://csvfit.com/blog/adversarial-validation https://csvfit.com/blog/adversarial-validation Thu, 23 Apr 2026 00:00:00 GMT A ten-line trick for finding data leaks and distribution shift: train a classifier to predict whether a row came from train or test, then read off which features it leans on. https://csvfit.com/blog/feature-importance-and-partial-dependency-plots https://csvfit.com/blog/feature-importance-and-partial-dependency-plots Sat, 18 Apr 2026 00:00:00 GMT Cracking open the black box: which inputs is your model actually using, and how does each one bend the prediction. Interactive widget, scribble diagrams, no Python. https://csvfit.com/blog/weighted-ensemble-l2 https://csvfit.com/blog/weighted-ensemble-l2 Sat, 18 Apr 2026 00:00:00 GMT What AutoGluon's WeightedEnsemble_L2 actually does: Caruana's 2004 greedy ensemble selection with replacement. Not L2 regularization — layer 2. Not sklearn — pure NumPy.