h2otools: Machine Learning Model Evaluation for 'h2o' Package

Model evaluation

There are plenty of procedures for evaluating machine learning models, many of which are not implemented in h2o platform. This repository provides additional functions for model performance evaluation that are not implemented in h2o.

The bootperformance function evaluates the model for n number of bootstrapped samples from the testing dataset, instead of evaluating the model on the testing dataset once. Therefore, evaluating the confidence interval of the model performance.

These functions are briefly described below:

Function	Description
automlModelParam	for extracting model parameters from AutoML grid
bootperformance	Bootstrap performance evaluation
Fmeasure	for evaluating F3, F4, F5, or any beta value. h2o only provides F0.5, F1, and F2
getPerfMatrix	retrieve performance matrix for all thresholds
kappa	Calculates kappa for all thresholds
performance	provides performance measures (AUC, AUCPR, MCC, Kappa, etc.) using objects from h2o package

Additional functions

There are plenty of procedures for evaluating machine learning models, many of which are not implemented in h2o platform. This repository provides additional functions for model performance evaluation that are not implemented in h2o.

The bootperformance function evaluates the model for n number of bootstrapped samples from the testing dataset, instead of evaluating the model on the testing dataset once. Therefore, evaluating the confidence interval of the model performance.

These functions are briefly described below:

Function	Description
checkFrame	Checks data.frame format, which is useful before uploading it to H2O cloud
h2o.get_ids	Extracts model IDs from h2o AutoML and Grids nd returns a vector of model IDs

Installation

You can install the latest stable package from CRAN:

install.packages("h2otools")