NN Classifier RL Agent Conv Vision Attention Playground

Everyone's AI

NN Classifier RL Agent Conv Vision Attention Playground

Playground

Neural Network Playground

Tinker with a neural network right here!

Epoch0

Learning rateActivationRegularizationReg. strength

Show formulas

Training settings help

These settings apply each time you play or step training. Try the defaults first, then change values and watch what happens.

Learning rate: How big each weight update is. Too large jumps around; too small learns slowly. Try around 0.01–0.03 to start.
Activation: Non-linear curve applied to each neuron’s output. Tanh is smooth (−1 to 1); ReLU zeros out negative values.
Regularization: Penalty to stop weights from growing too large. Helps reduce overfitting (memorizing the training set).
Reg. strength: How strong the regularization penalty is. Ignored when regularization is None.

Data

Which K-culture dataset?

Train/Test ratio 70%Noise 0.5Batch size 10

Show test data

Features

Choose inputs for the network

2/7

About features

A feature is an input value derived from each point's coordinates (x₁, x₂). Use raw X₁ and X₂, or enable squares, products, and sin terms so the model can learn more complex decision boundaries. Each enabled feature adds one input neuron.

Hidden layers

Line thickness = weight magnitude, color = sign (purple=+, orange=−)

H14

H22

Output

Model decision boundary and data points

Test loss

—

Training loss

—

Epoch: 0

Negative (−1)Positive (+1)

Faint background = true data pattern · bold color = network prediction