Live training dashboard
Results appear here while the tiny model learns.
Ready
Epoch
0
of 40
Loss
—
Lower is better
Train score
—
Current dataset
Test score
—
Unseen examples
Parameters
0
Tiny model size
Game score
0
☆☆☆
Training progress0%
Interactive learning playground
Drag examples, add labels, and inspect predictions directly.
Prediction inspector
Select a point to view its features, label, prediction, confidence, and error.
Tiny coach
Choose a preset, adjust settings, then train the model.
Plotly learning reports
Zoom, pan, hover, and export each interactive graph.
Prediction results
Sort, inspect, and export individual model decisions.
| # | Split | X | Y | Actual | Prediction | Confidence | Error | Status |
|---|---|---|---|---|---|---|---|---|
| Train a model to generate predictions. | ||||||||
Learn while playing
Open a concept for a simple explanation and practical hint.
The learning rate controls each update step. Tiny values learn slowly. Large values can overshoot useful solutions.
An epoch processes every training example once. Batches divide examples into smaller updates. Smaller batches often look noisier.
Underfitting means the model learned too little. Overfitting means training improved while unseen performance weakened. Validation curves expose both problems.
A decision boundary separates predicted classes. Straight boundaries suit simple patterns. Curved boundaries need hidden layers or transformed features.
Training log
Technical events from the latest run.
[ready] TensorFlow.js playground initialised.