Yet Another 'ARFF' Reader.

yarr

Yet Another ARFF Reader

So you need to read an ARFF file. You can use:

• foreign::read.arff, but only on dense format.
• farff::readARFF, but only on dense format.
• RWeka::read.arff, but you need to configure Java.

yarr::read.arff can read dense and sparse ARFF files, and it's implemented in pure R.

The implementations in R are derivatives of those in mldr and mldr.datasets, packages for management of multilabel learning datasets.

Usage

remotes::install_github("fdavidcl/yarr")
library(yarr)
download.file("https://www.openml.org/data/download/1681111/phpEUwA95", "dexter.arff")
dexter <- read.arff("dexter.arff")
dexter
# An ARFF dataset: dexter
# 20001 attributes and 600 instances
#   V0: numeric (min = 0, mean = 0.12, max = 72) 0, 0, 0, 0, 0...
#   V1: numeric (min = 0, mean = 0, max = 0) 0, 0, 0, 0, 0...
#   V2: numeric (min = 0, mean = 0, max = 0) 0, 0, 0, 0, 0...
#   V3: numeric (min = 0, mean = 1.165, max = 427) 0, 0, 0, 0, 0...
#   V4: numeric (min = 0, mean = 0.57, max = 342) 0, 0, 0, 0, 0...
#   V5: numeric (min = 0, mean = 0.386666666666667, max = 116) 0, 0, 0, 0, 0...
#   V6: numeric (min = 0, mean = 0, max = 0) 0, 0, 0, 0, 0...
#   V7: numeric (min = 0, mean = 0.448333333333333, max = 121) 0, 0, 0, 0, 0...
#   V8: numeric (min = 0, mean = 0.358333333333333, max = 84) 0, 0, 0, 0, 0...
#   ...
#   class: {-1,1} (-1: 300, 1: 300) 1, 1, 1, -1, 1...
relation(dexter)
# [1] "dexter"
write.arff(dexter, "dexter-new.arff")

Supported features

• Dense format
• Sparse format ({index value}, e.g. {0 1, 20 1})
• Missing values with ?
• MEKA parameter c in @relation.

Unsupported features

• Date attributes are read as mere strings
• Relational attributes are not supported
• Instance weights.