rate_decay

  • Available in: Deep Learning
  • Hyperparameter: yes

Description

The learning rate decay parameter controls the change of learning rate across layers. For example, assume the rate parameter is set to 0.01, and the rate_decay parameter is set to 0.5. Then the learning rate for the weights connecting the input and first hidden layer will be 0.01, the learning rate for the weights connecting the first and the second hidden layer will be 0.005, and the learning rate for the weights connecting the second and third hidden layer will be 0.0025, etc.

This parameter is only active when adaptive learning rate is disabled.

Example

  • r
  • python
library(h2o)
h2o.init()

# import the mnist datasets from the bigdata folder
train <- h2o.importFile("https://s3.amazonaws.com/h2o-public-test-data/bigdata/laptop/mnist/train.csv.gz")
test <- h2o.importFile("https://s3.amazonaws.com/h2o-public-test-data/bigdata/laptop/mnist/test.csv.gz")

 # Turn response into a factor (we want classification)
 train[,785] <- as.factor(train[,785])
 test[,785] <- as.factor(test[,785])
 train <- h2o.assign(train, "train")
 test <- h2o.assign(test, "test")

 # Train a deep learning model
 dl_model <- h2o.deeplearning(x=c(1:784), y=785,
                              training_frame=train,
                              activation="RectifierWithDropout",
                              adaptive_rate=F,
                              rate=0.01,
                              rate_decay=0.9,
                              rate_annealing=1e-6,
                              momentum_start=0.95,
                              momentum_ramp=1e5,
                              momentum_stable=0.99,
                              nesterov_accelerated_gradient=F,
                              input_dropout_ratio=0.2,
                              train_samples_per_iteration=20000,
                              classification_stop=-1,  # Turn off early stopping
                              l1=1e-5
                             )

 # See the model performance
 print(h2o.performance(dl_model, test))