README.md

In Richyrecarey/classifast: Automatic Testing And Stacking Of Classifiers For Predicting

The R package classifast is a side project that my friends Manuel, David and I are starting. Basically, the main goal is to make a function that, given multivariate labeled data, gives you a quick overview of which classifiers are the best for your problem (based on several accuracy criterious). Afterwards, the predict.classifast() method lets you predict using several techniques such as Stacking, weighted voting, etc.

This minimal package is intended to be a project to both enhance our R skills and to make (hopefully) a usefull tool for Data Scientists. All help and feedback will be highly appreciated, so feel free to contact me on [e-mail] to get involved in the project.

Instalation

install.packages(devtools) devtools::install_github("Richyrecarey/classifast")

Usage

Given multivariate data in a matrix-like objet x (rows are observations, columns are variables) and its correct labels in a vector-like object y, the main function classifast(x, y, method) will train the desired statistical classifiers selected in the string vector method. Several options are:

• method = "log": It will train binary or multinomial logistic regression, as needed.
• method = "svm": It will train Radial SVM.
• method = "knn": It will train the knn algorithm.
• method = "RandomForest": It will train Random Forest.
• method = "simple": It's the same as method = c("log", "svm", "knn", "RandomForest)
• More methods are under development

All the options and parameters available to tweak are available at help(classifast).

Once trained, classifast() will return an object of class classifast, which has several implemented methods/functions:

Implemented methods

summary.classifast()

# Train several classifiers on the Iris dataset
output <- classifast(x = iris[-5], y = iris[5], method = "simple")

# Show retults of chosen methods
summary(output)

And the output should look somehow like this:

 k-fold accuracy was approximated using 10-fold validation 

 Accuracy (%) of the diferent methods used: 

        Method     kf %   Test % Train %
1          log    96.66    92.45     100
2          svm    65.92    62.54    99.7
3          knn    75.51    72.54    94.3
4 RandomForest    84.72    82.54    92.4

predict.classifast()

Still developing

Richyrecarey/classifast documentation built on May 23, 2019, 1:09 p.m.