Active learning is a machine-learning paradigm for optimally choosing unlabeled observations in a training data set to query for their true labels. The framework is particularly useful when there are very few labeled observations relative to a large number of unlabeled observations, and the user seeks to determine as few true labels as possible to achieve highly accurate classifiers.
The activelearning R package is a collection of various active-learning
methods from the literature to optimally query observations with respect to a
variety of objective functions. Some active learning methods require posterior
probability estimates of the unlabeled observations from a single classifier or
a committee of classifiers; this package allows the user to specify custom
classifiers or specify a variety of classifiers by interfacing with the
caret R package.
The activelearning package implements the active learning methods as defined
by the excellent
literature survey
from Dr. Burr Settles. This literature
survey is also available in
book form,
which is highly recommended.
You can install the stable version on CRAN:
install.packages('activelearning', dependencies = TRUE)If you prefer to download the latest version, instead type:
library(devtools)
devtools::install_github('ramhiser/activelearning')