INTERACTIVE MODEL LAB

See how a model learns.

Change the experiment. Press play. Watch the boundary, loss and weights evolve.

REAL TRAINING · IN YOUR BROWSER

Decision boundary

Epoch 0

● Class 0● Class 1○ Held-out point

Class 0Class 1

The dark line is the learned boundary. Hover or tap the plot to inspect the network at that point.

Learning in progress

Training loss—

Held-out loss—

Held-out accuracy—

Parameters33

Learning curve

Cross-entropy ↓

Epochs

— Training— Held-out

Network connections

Ready to train.

Your comparison runs

Saved locally for this page session. Matching seeds and data settings make runs comparable.

Model / neurons	Data / seed	Settings	Epochs	Held-out loss	Accuracy
Train and save a result to begin.

Try these experiments

Linear versus nonlinear. Train the logistic classifier on circles, save the result, then switch to a neural network. Keep the data seed and split unchanged.

Noise and generalization. Increase label noise and compare training loss with held-out loss. A better fit to training data does not always mean better predictions.

Educational implementation inspired by TensorFlow Playground. Original implementation; no TensorFlow code or model weights used. Repeatedly inspecting held-out results can bias choices, so this lab is not a production evaluation protocol.