Awesome

JPMML-R

Java library and command-line application for converting R models to PMML.

Table of Contents

• Features
• Prerequisites
• Installation
• Usage
  • The R side of operations
  • The JPMML-R side of operations
• Documentation
• License
• Additional information

Features

• Fast and memory-efficient:
  • Can produce a 5 GB Random Forest PMML file in less than 1 minute on a desktop PC
• Supported model and transformation types:
  • ada package:
    • ada - Stochastic Boosting (SB) classification
  • adabag package:
    • bagging - Bagging classification
    • boosting - Boosting classification
  • apollo
    • apollo (formerly maxLik) - Discrete Choice Model (DCM) classification
  • caret package:
    • preProcess - Transformation methods "range", "center", "scale" and "medianImpute"
    • train - Selected JPMML-R model types
  • caretEnsemble package:
    • caretEnsemble - Ensemble regression and classification
  • CHAID package:
    • party - CHi-squared Automated Interaction Detection (CHAID) classification
  • earth package:
    • earth - Multivariate Adaptive Regression Spline (MARS) regression
  • elmNNRcpp package:
    • elm - Extreme Learning Machine (ELM) regression
  • evtree package:
    • party - Evolutionary Learning of Trees (EvTree) regression and classification
  • e1071 package:
    • naiveBayes - Naive Bayes (NB) classification
    • svm - Support Vector Machine (SVM) regression, classification and anomaly detection
  • gbm package:
    • gbm - Gradient Boosting Machine (GBM) regression and classification
  • glmnet package:
    • glmnet (elnet, fishnet, lognet and multnet subtypes) - Generalized Linear Model with lasso or elasticnet regularization (GLMNet) regression and classification
    • cv.glmnet - Cross-validated GLMNet regression and calculation
  • IsolationForest package:
    • iForest - Isolation Forest (IF) anomaly detection
  • lightgbm package:
    • lgb.Booster - LightGBM regression and classification.
  • MASS package:
    • negbin - Generalized Linear Model (GLM) regression.
  • mlr package:
    • WrappedModel - Selected JPMML-R model types.
  • neuralnet package:
    • nn - Neural Network (NN) regression
  • nnet package:
    • multinom - Multinomial log-linear classification
    • nnet.formula - Neural Network (NNet) regression and classification
  • party package:
    • ctree - Conditional Inference Tree (CIT) classification
  • partykit package:
    • party - Recursive Partytioning (Party) regression and classification
  • pls package:
    • mvr - Multivariate Regression (MVR) regression
  • pscl package:
    • hurdle - Hurdle regression
  • randomForest package:
    • randomForest - Random Forest (RF) regression and classification
  • ranger package:
    • ranger - Random Forest (RF) regression and classification
  • rms package:
    • lrm - Binary Logistic Regression (LR) classification
    • ols - Ordinary Least Squares (OLS) regression
  • rpart package:
    • rpart - Recursive Partitioning (RPart) regression and classification
  • r2pmml package:
    • scorecard - Scorecard regression
  • stats package:
    • glm - Generalized Linear Model (GLM) regression and classification:
      • binomial, gaussian, Gamma, inverse.gamma and poisson families
      • MASS::negative.binomial family
      • statmod::tweedie family
    • kmeans - K-Means clustering
    • lm - Linear Model (LM) regression
  • xgboost package:
    • xgb.Booster - XGBoost (XGB) regression and classification
• Data pre-processing using model formulae:
  • Interaction terms
  • base::I(..) function terms:
    • Logical operators &, | and !
    • Relational operators ==, !=, <, <=, >= and >
    • Arithmetic operators +, -, *, /, and %
    • Exponentiation operators ^ and **
    • The is.na function
    • Arithmetic functions abs, ceiling, exp, floor, log, log10, round and sqrt
  • base::cut() and base::ifelse() function terms
  • plyr::revalue() and plyr::mapvalues() function terms
• Production quality:
  • Complete test coverage.
  • Fully compliant with the JPMML-Evaluator library.

Prerequisites

• Java 1.8 or newer.

Installation

Enter the project root directory and build using Apache Maven:

mvn clean install

The build produces a library JAR file pmml-rexp/target/pmml-rexp-1.6-SNAPSHOT.jar, and an executable uber-JAR file pmml-rexp-example/target/pmml-rexp-example-executable-1.6-SNAPSHOT.jar.

Usage

A typical workflow can be summarized as follows:

1. Use R to train a model.
2. Serialize the model in RDS data format to a file in a local filesystem.
3. Use the JPMML-R command-line converter application to turn the RDS file to a PMML file.

The R side of operations

The following R script trains a Random Forest (RF) model and saves it in RDS data format to a file rf.rds:

library("randomForest")

rf = randomForest(Species ~ ., data = iris)

saveRDS(rf, "rf.rds")

The JPMML-R side of operations

Converting the RDS file rf.rds to a PMML file rf.pmml:

java -jar pmml-rexp-example/target/pmml-rexp-example-executable-1.6-SNAPSHOT.jar --rds-input rf.rds --pmml-output rf.pmml

Getting help:

java -jar pmml-rexp-example/target/pmml-rexp-example-executable-1.6-SNAPSHOT.jar --help

The conversion of large files (1 GB and beyond) can be sped up by increasing the JVM heap size using -Xms and -Xmx options:

java -Xms4G -Xmx8G -jar pmml-rexp-example/target/pmml-rexp-example-executable-1.6-SNAPSHOT.jar --rds-input rf.rds --pmml-output rf.pmml

Documentation

Up-to-date:

• Converting logistic regression models to PMML documents
• Deploying R language models on Apache Spark ML

Slightly outdated:

• Converting R to PMML

License

JPMML-R is licensed under the terms and conditions of the GNU Affero General Public License, Version 3.0.

If you would like to use JPMML-R in a proprietary software project, then it is possible to enter into a licensing agreement which makes JPMML-R available under the terms and conditions of the BSD 3-Clause License instead.

Additional information

JPMML-R is developed and maintained by Openscoring Ltd, Estonia.

Interested in using Java PMML API software in your company? Please contact info@openscoring.io