How to import/export to libsvm?

Question

Under the hood Mathematica uses libsvm, so I don't think this question is beyond the pale.

I need to export an svm I trained with Classify to the standard SVM Format (which is readable by libsvm).

The SVM format obeys the following Backus-Naur Form (see http://www.cs.cornell.edu/People/tj/svm_light/), it looks like this:

<line> ::= <target> " " (<feature> ":" <value>)+
<target> ::= <positive-integer>
<feature> ::= <positive-integer>
<value> ::= <float>

How can I transform the internal representation of Mathematica's Classifiers into this representation?

Notes:

• MATLAB provides libsvmread to import svms.
• In python sklearn.datasets.load_svmlight_file imports svms.
• Libsvm uses the so called "sparse" format where zero values do not need to be stored.

Answer 1

The problem is not able, but efficiency in training sparse data.

There are several purposes to import a libsvm dataset: one is training or predicting one is feature visualizing, doing something statistics and one is check debug data Of cause, there are many other purposes.

It's easy to write some functions and training them like IRIS, agaricus, lightweight datasets.

importSVMSimple[file_,n_:10]:=Block[{lines,data,label,data0},
lines=ReadList[file,"Record",n];
data=StringSplit[#,{" ",":"}]&/@lines;
label=First/@data;
data0=GroupBy[Partition[Rest@#,2],First->Last,#[[1]]&]&/@data;
Thread[data0->label]
]

iris=ExampleData[{"MachineLearning","FisherIris"}, "Data"];
iris[[1;;10]]

{{5.1,3.5,1.4,0.2}->setosa,{4.9,3.,1.4,0.2}->setosa,{4.7,3.2,1.3,0.2}->setosa,{4.6,3.1,1.5,0.2}->setosa,{5.,3.6,1.4,0.2}->setosa,{5.4,3.9,1.7,0.4}->setosa,{4.6,3.4,1.4,0.3}->setosa,{5.,3.4,1.5,0.2}->setosa,{4.4,2.9,1.4,0.2}->setosa,{4.9,3.1,1.5,0.1}->setosa}

data2LibSVM[dataSets_]:=Block[{labels,labelID,data},
labels=dataSets[[All,2]]//Union;
labelID=Association[Thread[labels->Range[Length@labels]]];
    data=MapAt[labelID,Map[MapIndexed[StringRiffle[{First@#2,#1},":"]&],dataSets,{2}],{All,2}];
StringRiffle[Flatten@Reverse@#," "]&/@List@@@data
]

Export["iris.libsvm.csv",data2LibSVM@iris]

This structure is trainable.

dataSet=importSVMSimple["iris.libsvm.csv",10000];
dataSet//Length
150
training=RandomSample[dataSet,100];
testing=Complement[dataSet,training];
model=Classify[training];
ClassifierMeasurements[model,testing,"Accuracy"]
0.95
model@testing[[1,1]]
1

---

We can write many related functions.