• Available in: GBM, DRF, GLM
  • Hyperparameter: no


Use this option to upload a custom metric function into a running H2O cluster. A custom metric function can be used to produce adhoc scoring metrics if actuals are presented.

Three separate fields must be specified when using this function:

  • klazz: Represents a custom function.
  • func_name: Assigns a name with uploaded custom functions. This name corresponds to the name of the key in the distributed key-value store.
  • func_file: The name of the file to store the function in an uploaded jar file. The source code of the given class is saved into a file that is subsequently zipped, uploaded as a zip-archive, and saved into the distributed key-value store.

The parameters func_name and func_file must be unique for each uploaded custom distribution.

Note: This option is only supported in the Python client.


import h2o
from h2o.estimators.gbm import H2OGradientBoostingEstimator

# import the airlines dataset:
# This dataset is used to classify whether a flight will be delayed 'YES' or not "NO"
# original data can be found at http://www.transtats.bts.gov/
airlines= h2o.import_file("https://s3.amazonaws.com/h2o-public-test-data/smalldata/airlines/allyears2k_headers.zip")

# convert columns to factors
airlines["Year"] = airlines["Year"].asfactor()
airlines["Month"] = airlines["Month"].asfactor()
airlines["DayOfWeek"] = airlines["DayOfWeek"].asfactor()
airlines["Cancelled"] = airlines["Cancelled"].asfactor()
airlines['FlightNum'] = airlines['FlightNum'].asfactor()

# set the predictor names and the response column name
predictors = ["Origin", "Dest", "Year", "UniqueCarrier", "DayOfWeek", "Month", "Distance", "FlightNum"]
response = "IsDepDelayed"

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

# try using the `stopping_metric` parameter:
# since this is a classification problem we will look at the AUC
# you could also choose logloss, or misclassification, among other options
# train your model, where you specify the stopping_metric, stopping_rounds,
# and stopping_tolerance
# initialize the estimator then train the model
airlines_gbm = H2OGradientBoostingEstimator(stopping_metric="auc",
airlines_gbm.train(x=predictors, y=response, training_frame=train, validation_frame=valid)

# print the auc for the validation data

# Use a custom metric
# Create a custom RMSE Model metric and save as mm_rmse.py
# Note that this references a java class java.lang.Math
class CustomRmseFunc:
def map(self, pred, act, w, o, model):
    idx = int(act[0])
    err = 1 - pred[idx + 1] if idx + 1 < len(pred) else 1
    return [err * err, 1]

def reduce(self, l, r):
    return [l[0] + r[0], l[1] + r[1]]

def metric(self, l):
    # Use Java API directly
    import java.lang.Math as math
    return math.sqrt(l[0] / l[1])

# Upload the custom metric
custom_mm_func = h2o.upload_custom_metric(CustomRmseFunc,

# Train the model
model = H2OGradientBoostingEstimator(ntrees=3,
model.train(x=predictors, y=response, training_frame=train, validation_frame=valid)