rvwCpp

Test R interface to Vowpal Wabbit for GSoC 2018.

Written using Rcpp and libvw.

Requirements

This package requires vw libraries. You can get them here.

vw libraries should be installed in default directories: /usr/local/include/, /usr/local/lib

Also requires perf program to compute AUC score.

Installation

library(devtools)
install_github("ivan-pavlov/rvwCpp")

Examples

Demonstration file is in demo/rvwCpp_example.R

Function vwCpp uses files in .vw format.

aucCpp <- vwCpp(training_data = "diamond_train.vw", validation_data = "diamond_val.vw", 
			validation_labels = "valid_labels.txt", out_probs = "predsCpp.vw", model = "mdlCpp.vw",
			loss = "logistic", link_function = "logistic", do_evaluation = TRUE,
			b = 25, learning_rate = 0.5, passes = 1)

Usage

Parameters	Description
training_data	File in vw format with data to train model
validation_data	File in vw format with data to compute predictions
validation_labels	plain text file with true labels from validation_data. One label on each line
out_probs	File where to write final predictions
model	File where to write final model
loss	loss function, default is "logistic"
link_function	link function, default is "logistic"
b	Number of bits in the feature table, default is 25
learning_rate	Initial learning_rate, default is 0.5
passes	Number of training passes, default is 1
do_evaluation	If TRUE, AUC score will be computed using perf