step!
function defined in module
FluxTraining
step!(learner, phase::Phase, batch)
Run one step of training for
learner on batch. Behavior is customized through
phase.
This is a required method for custom
Phase
s to implement. To implement
step!, it is recommended you make use of
runstep to get begin and end events as well as proper handling of
CancelStepExceptions.
See the implementations of
TrainingPhase and
ValidationPhase for reference.
There are
5
methods for FluxTraining.step!:
The following pages link back here:
Performant data pipelines, Variational autoencoders
training/lrfind.jl , FluxTraining.jl , callbacks/metrics.jl , training.jl