OCAPIS [DEPRECATED]

Package for ordinal data classification and preprocessing implementing algorithms in Scala

---

Included algorithms are:

Classification algorithms

• Support vector machine for ordinal data
• Ordinal regression
• Kernel Discriminant Learning for Ordinal Regression
• Weighted K-Nearest Neightborgs for monotonic and ordinal data

Preprocessing algorithms

• Feature selection for monotonic and ordinal data
• Instance selection for monotonic and ordinal data

Installation

Dependencies

Before installing OCAPIS you need to get Python (> 2.7), Scala(>=2.11) and libsvm-weights-3.17 installed on your system, if they are not yet.

Installing Python

If using Linux, you can easily install Python from the command line, just typing:

$ sudo apt-get install python3

If your system is an Ubuntu distribution, or its counterpart in the distro you use. If you are not using Linux or you are not convinced to install Python through command line, just check this official Python Installation guide.

Installing Scala

Similarly, if using Linux, you can install Scala from repo. For example, for Linux Mint just type:

sudo apt install scala

In any other case just check the Other ways to install Scala section from the official Scala Installation Guide.

Installing Libsvm-weights

Libsvm-weights-3.17 is required as it is used by SVMOP method. Please , keep in mind that OCAPIS will look for the libsvm-weights in the following directories:

If you are using MAC or Linux:

• "/usr/lib/libsvm-weights-3.17/python"
• "/usr/local/lib/libsvm-weights-3.17/python"

If you are using Windows:

• "C:\Program Files (x86)/libsvm-weights-3.17/python"
• "C:\Program Files (x86)/libsvm-weights-3.17/python"

So, just extract the downloaded libsvm-weights-3.17 folder in one of the above paths and compile it following the instructions in Installation and Data Format section from the README on Libsvm-weights-3.17 or Libsvm-weights-3.17 gitHub.

For example, as I am using Linux, I have:

cris@cris /usr/local/lib/libsvm-weights-3.17 $ ls
 COPYRIGHT  heart_scale      libsvm.so.2  Makefile.win  python  README.weight  svm.def  svm.o          svm-scale.c  tools
 FAQ.html   heart_scale.wgt  Makefile     matlab        README  svm.cpp        svm.h    svm-predict.c  svm-train.c

and inside /python, I have the .py files:

cris@cris /usr/local/lib/libsvm-weights-3.17/python $ ls
 Makefile  README  README.weight  svm.py  svm.pyc  svmutil.py  svmutil.pyc

Installing OCAPIS

After installing the external dependencies, the latest version of OCAPIS can be installed from GitHub via:

devtools::install_github("cristinahg/OCAPIS/OCAPIS")

The rest of the dependencies will be automatically installed. These are Reticulate and Rscala.

Usage

Below are shown examples of how to use all classification and preprocessing methods, using an ordinal dataset named balance-scale.

Classification

# Data reading
dattrain<-read.table("train_balance-scale.0", sep=" ")
trainlabels<-dattrain[,ncol(dattrain)]
traindata=dattrain[,-ncol(dattrain)]
dattest<-read.table("test_balance-scale.0", sep=" ")
testdata<-dattest[,-ncol(dattest)]
testlabels<-dattest[,ncol(dattest)]

# SVMOP
modelstrain<-svmofit(traindata,trainlabels,TRUE,0.1,0.1)
predictions<-svmopredict(modelstrain,testdata)
sum(predictions[[2]]==testlabels)/nrow(dattest)

# POM
fit<-pomfit(traindata,trainlabels,"logistic")
predictions<-pompredict(fit,testdata)
projections<-predictions[[1]]
predictedLabels<-predictions[[2]]
sum(predictedLabels==testlabels)/nrow(dattest)

# KDLOR
myfit<-kdlortrain(traindata,trainlabels,"rbf",10,0.001,1)
pred<-kdlorpredict(myfit,traindata,testdata)
sum(pred[[1]]==testlabels)/nrow(dattest)

# WKNNOR
predictions<-wknnor(traindata,trainlabels,testdata,5,2,"rectangular",FALSE)
sum(predictions==testlabels)/nrow(dattest)
mae(testlabels,predictions)

Preprocessing

# Feature Selector
selected<-fselector(traindata,trainlabels,2,2,8)
trainselected<-traindata[,selected]

# Instance Selector
selected<-iselector(traindata,trainlabels,0.02,0.1,5)
trainselected<-selected[,-ncol(selected)]
trainlabels<-selected[,ncol(selected)]

For more details about method params, see OCAPIS documentation.

References:

1. E. Frank and M. Hall, "A simple approach to ordinal classification" in Proceedings of the 12th European Conference on Machine Learning, ser. EMCL'01. London, UK: Springer-Verlag, 2001, pp. 145–156. https://doi.org/10.1007/3-540-44795-4_13
2. W. Waegeman and L. Boullart, "An ensemble of weighted support vector machines for ordinal regression", International Journal of Computer Systems Science and Engineering, vol. 3, no. 1, pp. 47–51, 2009.
3. P.A. Gutiérrez, M. Pérez-Ortiz, J. Sánchez-Monedero, F. Fernández-Navarro and C. Hervás-Martínez Ordinal regression methods: survey and experimental study IEEE Transactions on Knowledge and Data Engineering, Vol. 28. Issue 1 2016 http://dx.doi.org/10.1109/TKDE.2015.2457911
4. P. McCullagh, Regression models for ordinal data, Journal of the Royal Statistical Society. Series B (Methodological), vol. 42, no. 2, pp. 109–142, 1980.
5. B.-Y. Sun, J. Li, D. D. Wu, X.-M. Zhang, and W.-B. Li, Kernel discriminant learning for ordinal regression IEEE Transactions on Knowledge and Data Engineering, vol. 22, no. 6, pp. 906-910, 2010. https://doi.org/10.1109/TKDE.2009.170
6. Duivesteijn, Wouter & Feelders, Ad. (2008). Nearest Neighbour Classification with Monotonicity Constraints. 301-316. 10.1007/978-3-540-87479-9_38.
7. Cano, José & García, S. (2017). Training Set Selection for Monotonic Ordinal Classification. Data & Knowledge Engineering. 112. 10.1016/j.datak.2017.10.003.
8. Hu, Qinghua & Pan, Weiwei & Zhang, Lei & Zhang, David & Song, Yanping & Guo, Maozu & Yu, Daren. (2012). Feature Selection for Monotonic Classification. IEEE T. Fuzzy Systems. 20. 69-81. 10.1109/TFUZZ.2011.2167235.
9. Hechenbichler, Schliep: Weighted k-Nearest-Neighbor Techniques and Ordinal Classification Sonderforschungsbereich 386, Paper 399 (2004) https://epub.ub.uni-muenchen.de/1769/1/paper_399.pdf