lre_min

  • Available in: CoxPH

  • Hyperparameter: no

Description

The lre_min option measures the relative difference of log likelihood before and after iteration of the CoxPH algorithm. When building a CoxPH model, the algorithm stops when \(|(logLik - newLoglik) / newLoglik| <= 1e-9\). This value defaults to 9.

Example

library(h2o)
h2o.init()

# import the heart dataset
heart <- h2o.importFile("http://s3.amazonaws.com/h2o-public-test-data/smalldata/coxph_test/heart.csv")

# split the dataset into train and test datasets
heart_split <- h2o.splitFrame(data = heart, ratios = 0.8, seed = 1234)
train <- heart_split[[1]]
test <- heart_split[[2]]

# train a CoxPH model
coxph_model <- h2o.coxph(x = "age",
                         event_column = "event",
                         start_column = "start",
                         stop_column = "stop",
                         training_frame = heart,
                         lre_min = 9)

# run prediction against the test dataset
predicted <- h2o.predict(coxph_model, test)

# view the predictions
predicted
           lp
1 0.26964730
2 0.16438761
3 0.07569035
4 0.27813870
5 0.27813870
6 0.26090368

[34 rows x 1 column]
    import h2o
    from h2o.estimators.coxph import H2OCoxProportionalHazardsEstimator
    h2o.init()

    # import the heart dataset
    heart = h2o.import_file("http://s3.amazonaws.com/h2o-public-test-data/smalldata/coxph_test/heart.csv")

    # split the dataset into train and test datasets
    train, test = heart.split_frame(ratios = [.8], seed=1234)

    # set the lre_mind parameter's value
    lre_min = 9

    # initialize an train a CoxPH model
    coxph = H2OCoxProportionalHazardsEstimator(start_column="start",
                                               stop_column="stop",
                                               ties="breslow",
                                               lre_min=lre_min)
    coxph.train(x="age", y="event", training_frame=heart)

    # run prediction against the test dataset
    pred = coxph.predict(test_data=test)

    # view the predictions
    pred
       lp
---------
0.269501
0.164298
0.0756492
0.277987
0.277987
0.260762
0.260762
0.254712
0.347814
0.299666

[34 rows x 1 column]