h2o (version 2.8.4.4)

h2o.performance: Performance Measures

Description

Evaluate the predictive performance of a model via various measures.

Usage

h2o.performance(data, reference, measure = "accuracy", thresholds, gains = TRUE, ...)

Arguments

data

An H2OParsedData object containing the predicted outcome scores. Must be a single column with the same number of rows as reference.

reference

An H2OParsedData object containing the actual outcomes for comparison. Must be a single binary column with all entries in {0,1}.

measure

A character string indicating the performance measure to optimize. Must be one of the following:

• F1: F1 score, equal to$2*(Precision*Recall)/{Precision + Recall}$

accuracy: Accuracy of model, esti

Value

• An object of class H2OPerfModel with slots cutoffs, measure, perf (the performance measure selected), roc (data frame used to plot ROC) and model, where the last is a list of the following components:
• auc
  Area under the curve.
• gini
  Gini coefficient.
• best_cutoff
  Threshold value that optimizes the performance measure.
• F1
  F1 score at best cutoff.
• accuracy
  Accuracy value at best cutoff.
• precision
  Precision value at best cutoff.
• recall
  Recall value at best cutoff.
• specificity
  Specificity value at best cutoff.
• max_per_class_err
  Maximum per class error at best cutoff.
• confusion
  Confusion matrix at best cutoff.

item

• thresholds
• gains
• ...

Examples

library(h2o)
localH2O = h2o.init()

# Run GBM classification on prostate.csv
prosPath = system.file("extdata", "prostate.csv", package = "h2o")
prostate.hex = h2o.importFile(localH2O, path = prosPath, key = "prostate.hex")
prostate.gbm = h2o.gbm(y = 2, x = 3:9, data = prostate.hex)

# Calculate performance measures at threshold that maximizes precision
prostate.pred = h2o.predict(prostate.gbm)
head(prostate.pred)
h2o.performance(prostate.pred[,3], prostate.hex$CAPSULE, measure = "precision")