find_lr: Plot learning rate

In mlr-org/mlr3keras: mlr3 Keras extension

Plot learning rate

Description

Trains the model for few epochs, iteratively increasing learning rate. Tries to provide insight with respect to the choice of a learning rate by iteratively increasing learning rates from lr_min and lr_max, recording performance gains (c.f. Leslie N. Smith: Cyclical Learning Rates for Training Neural Networks 2015).

'find_lr' is also available via Learner$lr_find(...).

Usage

find_lr(
  learner,
  task,
  epochs = 5,
  lr_min = 10^-4,
  lr_max = 0.8,
  batch_size = 128L
)

Arguments

learner	Learner An mlr3 Learner from mlr3keras.
task	Task An mlr3 Task.
epochs	integer Number of epochs to train for. Defaults to 5.
lr_min	numeric Minimum learning rate to try. Defaults to 1e-4
lr_max	numeric Maximum learning rate to try. Defaults to 0.8
batch_size	numeric Batch size. Defaults to 128.

mlr-org/mlr3keras documentation built on April 12, 2022, 11:35 a.m.