Machine Learning algorithm: Testing a K-nearest neighbor algorithm

Before writing the K-nearest neighbor algorithm (http://boytnt.blog.51cto.com/966121/1569629), the test data is not attached, this time to find a, test the effect of the algorithm. Data from http://archive.ics.uci.edu/ml/machine-learning-databases/breast-cancer-wisconsin/ Breast-cancer-wisconsin.data, a sample of breast cancer, attribute description see Breast-cancer-wisconsin.names.

The approximate form of the sample is as follows:

1000025,5,1,1,1,2,1,3,1,1,2

The 1th attribute is number, we don't care, the last attribute is the result, 2 means Benign ( benign), 4 indicates Malignant ( malignant). The remaining 9 properties are sample features. Notice that there is missing data in it. Said, a total of 16 lines, accounting for 2.3%), the calculation of the first to do data cleaning, here simple filling into 0 can.

Use the K-nearest neighbor algorithm to test:

Public void testnearestneighbour () {    var trainingset = new  List<DataVector<double>> ();    var testset = new  List<datavector<double>> ();         //Read Data      var file = new streamreader ("Breast-cancer-wisconsin.txt",  Encoding.default);     for (int i = 0;i < 699;++i)      {        string line = file. ReadLine ();         var parts = line. Split (', ');                 var  p = new DataVector<double> (9);         for ( INT&NBSP;J&NBSP;=&NBSP;0;J&NBSP;&LT;&NBSP;P.DIMENSION;++J)   &nbsP;     {            if ( parts[j + 1] ==  "?")                 parts[j +  1] =  "0";             p.data[j]  = convert.todouble (parts[j + 1]);        }         p.label = convert.toint32 (parts[10])  == 2 ?   "Benign"  :  "malignant";                 //used 600 samples for training, leaving 99 to do test         if (i  < 600)             trainingset.add (p);         else       &nbsP;    testset.add (P);     }    file. Close ();         //Inspection     var nn =  New nearestneighbour ();     nn. Train (Trainingset);     int error = 0;    foreach (var  p in testset)     {        var  label = nn. Classify (P);         if (Label != p.label)              ++error;    }         console.writeline ("error = {0}/{1}, {2}%", error,  testset.count,  (Error * 100.0 / testset.count));}

The result is 99 test samples wrong 2, error rate 2.02%, the effect is good.

This article is from the "Rabbit Nest" blog, please be sure to keep this source http://boytnt.blog.51cto.com/966121/1572149

Machine Learning algorithm: Testing a K-nearest neighbor algorithm