Learn what your models see. Explore datasets, tune algorithms, and inspect predictions—entirely in your browser.
LIVE EXPERIMENT
Decision landscape
Generated points and model predictions
Not trained
Select a point to inspect · X₁ vs X₂
0 data pointsTraining uses a held-out test split
03
Explore the dataset
◎
Choose a plotted point or table row to inspect and edit it.
Preview of first 60 rows. Download CSV for all rows.
#
X₁
X₂
Target / class
Split
Page 1
Using Machine Learning Playground
Experiment with two-feature clustering, classification, and regression, then compare held-out metrics and training-only validation folds.
Workflow
Choose a generated example or import a two-feature CSV/TSV dataset, then select clustering, classification, or regression.
Choose an available algorithm, set the primary train/test split and feature standardization, then train and inspect the chart and metrics.
Run repeated validation folds using training rows only, compare predictions, and export results or an editable project backup.
Worked example
Load the labeled three-class example, select k-nearest neighbors, and train with the held-out test split. Inspect the decision landscape and held-out metrics, then run cross-validation on the training portion before comparing a second algorithm.
Limits and data handling
The teaching algorithms are bounded to two features and include K-means, k-NN, binary logistic regression, and linear/polynomial ridge regression. Held-out test rows are not used to fit or compare algorithms; validation folds come only from training rows. These experiments do not establish production model quality or suitability for consequential decisions. Standard ML training runs without model downloads. Optional local AI explanations download a language model and receive bounded aggregate statistics, not raw CSV rows or user class names; AI output can be wrong and requires local device resources.
Frequently asked questions
Does cross-validation include held-out test rows?
No. Repeated validation folds are drawn from the primary training rows; the held-out test data is kept separate from fitting and algorithm comparisons.
Do the clustering and regression algorithms require a model download?
No. Standard ML training runs in the browser without a pretrained model download. The separate optional AI explanation feature downloads a local language model and receives bounded aggregate statistics rather than raw rows or class names.
