getting_started.md

Getting started

Let's look at a simple training example. In FluxTraining.jl, a Learner holds all state necessary for training. To get started, you need

• a model
• training and validation data iterators
• a loss function; and
• an optimizer

First we define the necessary pieces:

model = ...
traindata, valdata = ...
lossfn = Flux.Losses.mse
opt = Flux.ADAM(0.01)

Then we construct a Learner:

learner = Learner(model, lossfn; data=(traindata, valdata))

And train for 10 epochs:

fit!(learner, 10)