Skip to main content

`max_after_balance_size`

  • Available in: GBM, DRF, Deep Learning, GLM, Naïve-Bayes, AutoML, Uplift DRF
  • Hyperparameter: yes

Description​

When your datasest includes imbalanced data, you may find it necessary to balance the data using the balance_classes option. When specified, the algorithm will either undersample the majority classes or oversampling the minority classes. In most cases, though, enabling the balance_classes option will increase the data frame size. To reduce the data frame size, you can use the max_after_balance_size option. This specifies the maximum relative size of the training data after balancing class counts and defaults to 5.0.

Example​

library(h2o)
h2o.init()

# import the covtype dataset: 
# this dataset is used to classify the correct forest cover type 
# original dataset can be found at https://archive.ics.uci.edu/ml/datasets/Covertype
covtype <- h2o.importFile("https://s3.amazonaws.com/h2o-public-test-data/smalldata/covtype/covtype.20k.data")

# convert response column to a factor
covtype[, 55] <- as.factor(covtype[, 55])

# set the predictor names and the response column name
predictors <- colnames(covtype[1:54])
response <- 'C55'

# split into train and validation sets
covtype_splits <- h2o.splitFrame(data =  covtype, ratios = 0.8, seed = 1234)
train <- covtype_splits[[1]]
valid <- covtype_splits[[2]]

# look at the frequencies of each class
print(h2o.table(covtype['C55']))

# try using the max_after_balance_size parameter:
max <- 0.85
cov_gbm <- h2o.gbm(x = predictors, y = response, training_frame = train,
                   validation_frame = valid, balance_classes = TRUE, 
                   max_after_balance_size = max, seed = 1234)

# print the logloss for your model
print(h2o.logloss(cov_gbm, valid = TRUE))

# grid over `max_after_balance_size`
# select the values for `max_after_balance_size` to grid over
# the first and last max_after_balance_sizes reduce the size of the
# original dataset, the second increases the dataset by 1.7 
hyper_params <- list( max_after_balance_size = c(0.85, 1.7, 0.5) )

# this example uses cartesian grid search because the search space is small
# and we want to see the performance of all models. For a larger search space use
# random grid search instead: {'strategy': "RandomDiscrete"}

# build grid search with previously made GBM and hyperparameters
grid <- h2o.grid(x = predictors, y = response, training_frame = train, validation_frame = valid,
                 algorithm = "gbm", grid_id = "covtype_grid", balance_classes =  TRUE, hyper_params = hyper_params,
                 search_criteria = list(strategy = "Cartesian"), seed = 1234)  

# Sort the grid models by logloss
sorted_grid <- h2o.getGrid("covtype_grid", sort_by = "logloss", decreasing = FALSE)    
sorted_grid
Feedback