From David's Wiki
Jump to navigation Jump to search

Flux is a machine learning library for Julia


Basic Usage

using Flux;
using Flux.Tracker: update!;

// Create layers like this
inputSize = 10;
outputSize = 20;
myLayer = Dense(inputSize, outputSize, relu);
// You can get the weights using
// You can call the model to make a prediction
// Equivalent to 
relu(model.W * myData + model.b);

// Create Networks like this
model = Chain(
  Dense(10, 20, relu),
  Dense(20, 2)

// Calling the model will pass data through each layer.

// Get the parameters for the whole model with
p = params(model);

// Calculate the gradient with
gs = Tracker.gradient(function()
  predicted = model(myData);
  loss = sum((predicted - myLabels).^2)'
  return loss;
end, params(model));
// Gradient of layer 1 weights
// Update! will update the weights and clear the gradient
// Make sure to update all layers.
update!(model[1].W, -0.1 * gs[model[1].W]);

// You can also define an optimizer and update using the optimizer
opt = Adam()
update!(opt, model[1].W, gs[model[1].W])