huber_alpha

  • Available in: GBM, Deep Learning
  • Hyperparameter: yes

Description

The Huber loss function is a combination of the squared-error loss function and absolute-error loss function. It applies the squared-error loss for small deviations from the actual response value and the absolute-error loss for large deviations from the actual respone value. To activate this parameter you must set distribution=huber and specify the huber_alpha parameter, which dictates the threshold between quadratic and linear loss (i.e. the top percentile of error that should be considered as outliers). This value must be between 0 and 1 and defaults to 0.9.

More information about the Huber loss function is available here.

Example

library(h2o)
h2o.init()

# import the insurance dataset:
# this dataset predicts the number of claims a policy holder will make
# original dataset can be found at https://cran.r-project.org/web/packages/MASS/MASS.pdf
insurance <- h2o.importFile("https://s3.amazonaws.com/h2o-public-test-data/smalldata/glm_test/insurance.csv")

# set the predictor names and the response column name
predictors <- colnames(insurance)[1:4]
response <- 'Claims'

# convert columns to factors
insurance['Group'] <- as.factor(insurance['Group'])
insurance['Age'] <- as.factor(insurance['Age'])

# split into train and validation sets
insurance.splits <- h2o.splitFrame(data =  insurance, ratios = .8, seed = 1234)
train <- insurance.splits[[1]]
valid <- insurance.splits[[2]]

# try using the `huber_alpha` parameter:
# train your model, where you specify the distribution as huber
# and the huber_alpha parameter
insurance_gbm <- h2o.gbm(x = predictors, y = response, training_frame = train,
                      validation_frame = valid,
                      distribution = 'huber',
                      huber_alpha = .9,
                      seed = 1234)

# print the MSE for validation set
print(h2o.mse(insurance_gbm, valid = TRUE))

# grid over `huber_alpha` parameter
# select the values for `huber_alpha` to grid over
hyper_params <- list( huber_alpha = c(.2, .5, .8) )

# 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 = "insurance_grid",
                 distribution = "huber",
                 hyper_params = hyper_params,
                 seed = 1234)

# Sort the grid models by MSE
sortedGrid <- h2o.getGrid("insurance_grid", sort_by = "mse", decreasing = FALSE)
sortedGrid
import h2o
from h2o.estimators.gbm import H2OGradientBoostingEstimator
h2o.init()

# import the insurance dataset:
# this dataset predicts the number of claims a policy holder will make
# original dataset can be found at https://cran.r-project.org/web/packages/MASS/MASS.pdf
insurance = h2o.import_file("https://s3.amazonaws.com/h2o-public-test-data/smalldata/glm_test/insurance.csv")

# set the predictor names and the response column name
predictors = insurance.columns[0:4]
response = 'Claims'

# convert columns to factors
insurance['Group'] = insurance['Group'].asfactor()
insurance['Age'] = insurance['Age'].asfactor()

# split into train and validation sets
train, valid = insurance.split_frame(ratios = [.8], seed = 1234)

# try using the `huber_alpha` parameter:
# initialize your estimator where you specify the distribution as huber
# and the huber_alpha parameter
insurance_gbm = H2OGradientBoostingEstimator(distribution="huber", huber_alpha = 0.9, seed =1234)

# then train your model
insurance_gbm.train(x = predictors, y = response, training_frame = train, validation_frame = valid)

# print the MSE for the validation data
print(insurance_gbm.mse(valid = True))


# Example of values to grid over for `huber_alpha`
# import Grid Search
from h2o.grid.grid_search import H2OGridSearch

# select the values for `huber_alpha` to grid over
hyper_params = {'huber_alpha': [.2, .5, .8]}

# 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"}
# initialize the GBM estimator
insurance_gbm_2 = H2OGradientBoostingEstimator(distribution="huber", seed = 1234)

# build grid search with previously made GBM and hyper parameters
grid = H2OGridSearch(model = insurance_gbm_2, hyper_params = hyper_params,
                     search_criteria = {'strategy': "Cartesian"})

# train using the grid
grid.train(x = predictors, y = response, training_frame = train, validation_frame = valid)

# sort the grid models by decreasing MSE
sorted_grid = grid.get_grid(sort_by = 'mse', decreasing = False)
print(sorted_grid)