TensorFlow.js playground

AI Model Explorer

Choose a data source, train a small neural network in your browser, then inspect how numeric features move through the model.

Loading dataset...Model not trained yetPrediction waits for training

What the model sees

Palmer Penguins inputs

Model predictionTrain first

A tabular classifier: each animal becomes measurements, and the model predicts a species.

Confidence

Output probabilities

Tensor nodes

How values move through the neural network

Circle size shows activation strength. To keep the graph readable, it draws the strongest active connections instead of every weight in the network.

input tensordense layer 1dense layer 2softmax outputBLBill length mmBDBill depth mmFLFlipper length mmBMBody mass g1H1.1waiting2H1.2waiting3H1.3waiting4H1.4waiting5H1.5waiting6H1.6waiting7H1.7waiting8H1.8waiting9H1.9waiting10H1.10waiting1H2.1waiting2H2.2waiting3H2.3waiting4H2.4waiting5H2.5waiting6H2.6waiting7H2.7waiting8H2.8waiting

Learning

Loss should fall, accuracy should rise

Training progresslossaccuracy

Data shape

Palmer Penguins examples

Errors

Confusion matrix

Train the model to see which labels it confuses.

Pluggable data source

What changed?

Each source defines how to fetch rows, which numeric features to use, and what label the model should learn.

The same TensorFlow model trains on penguins, flowers, insect records, or word features because they all become tensors with numeric columns.