“A Neural Network Playground”, Daniel Smilkov, Shan Carter2016-11-05 (neural net, JS, data visualization; similar)⁠:

[(Github) An in-browser implementation using TypeScript of a simple feedforward MLP, whose architecture, LR, activation, regularization, and task can be varied and the NN retrained with the intermediate function of each neuron visualized and the decision boundary on the data plotted.

One can see how different hyperparameters lead to different learned units and boundaries of varying smoothness and shapes, and how SGD updates it each iteration.

Available settings:

• Show test data

• Discretize output

• Play button

• Step button

• Reset button

• Learning rate

• Activation

• Regularization

• Regularization rate

• Problem type

• Which dataset

• Ratio train data

• Noise level

• Batch size

• number of hidden layers]

Similar Links: