Классифицировать и прогнозировать множественные атрибуты с помощью Weka

Question

Мне нужно ввести 6 атрибутов и классифицировать/предсказать 3 атрибута из этого ввода с помощью программ Java/Weka программно. Я выяснил, как предсказать 1 (последний) атрибут, но как я могу изменить это, чтобы обучать и прогнозировать последние 3 одновременно?

Номера в файлах .arff соответствуют объектам фильма в базе данных.

Вот мой Java-код:

import java.io.BufferedReader; 
import java.io.FileReader; 

import weka.classifiers.meta.FilteredClassifier; 
import weka.classifiers.trees.DecisionStump; 
import weka.classifiers.trees.J48; 
import weka.classifiers.trees.RandomForest; 
import weka.classifiers.trees.RandomTree; 
import weka.core.Instances; 
import weka.filters.unsupervised.attribute.Remove; 

public class WekaTrial { 

    /** 
    * @param args 
    * @throws Exception 
    */ 
    public static void main(String[] args) throws Exception { 

     // Create training data instance 
     Instances training_data = new Instances(
       new BufferedReader(
         new FileReader(
           "C:/Users/Me/Desktop/File_Project/src/movie_training.arff"))); 
     training_data.setClassIndex(training_data.numAttributes() - 1); 

     // Create testing data instance 
     Instances testing_data = new Instances(
       new BufferedReader(
         new FileReader(
           "C:/Users/Me/Desktop/FileProject/src/movie_testing.arff"))); 
     testing_data.setClassIndex(training_data.numAttributes() - 1); 

     // Print initial data summary 
     String summary = training_data.toSummaryString(); 
     int number_samples = training_data.numInstances(); 
     int number_attributes_per_sample = training_data.numAttributes(); 
     System.out.println("Number of attributes in model = " 
       + number_attributes_per_sample); 
     System.out.println("Number of samples = " + number_samples); 
     System.out.println("Summary: " + summary); 
     System.out.println(); 

     // a classifier for decision trees: 
     J48 j48 = new J48(); 

     // filter for removing samples: 
     Remove rm = new Remove(); 
     rm.setAttributeIndices("1"); // remove 1st attribute 

     // filtered classifier 
     FilteredClassifier fc = new FilteredClassifier(); 
     fc.setFilter(rm); 
     fc.setClassifier(j48); 

        // Create counters and print values 
     float correct = 0; 
     float incorrect = 0; 

        // train using stock_training_data.arff: 
     fc.buildClassifier(training_data); 

        // test using stock_testing_data.arff: 
     for (int i = 0; i < testing_data.numInstances(); i++) { 
      double pred = fc.classifyInstance(testing_data.instance(i)); 
      System.out.print("Expected values: " 
        + testing_data.classAttribute().value(
          (int) testing_data.instance(i).classValue())); 
      System.out.println(", Predicted values: " 
        + testing_data.classAttribute().value((int) pred)); 
      // Increment correct/incorrect values 
      if (testing_data.classAttribute().value(
        (int) testing_data.instance(i).classValue()) == testing_data.classAttribute().value((int) pred)) { 
         correct += 1; 
        } else { 
         incorrect += 1; 
        } 
     } 

        // Print correct/incorrect 
     float percent_correct = correct/(correct+incorrect)*100; 
     System.out.println("Number correct: " + correct + "\nNumber incorrect: " + incorrect + "\nPercent correct: " + 
       percent_correct + "%"); 

    } 

} 

Это мой файл обучения .arff (с лишние строки удалены):

@relation movie_data 

@attribute movie1_one {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie1_two {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie1_three {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie2_one {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie2_two {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie2_three {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute decision_one {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute decision_two {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute decision_three {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 

@data 
18,18,18,18,18,18,18,18,18 
28,18,36,18,53,10769,18,53,10769 
37,37,37,28,12,14,28,12,14 
27,53,27,18,10749,10769,27,53,27 
12,12,12,35,10751,35,12,12,12 
35,18,10749,18,18,18,35,18,10749 
28,12,878,53,53,53,53,53,53 
18,18,18,28,37,10769,18,18,18 
18,53,18,28,12,35,18,53,18 
28,80,53,80,18,10749,28,80,53 
18,10749,18,18,10756,18,18,10756,18 
18,10749,10769,28,12,878,18,10749,10769 
18,10756,18,16,35,10751,16,35,10751 
35,18,10751,35,18,10752,35,18,10751 

И файл тестирования .arff:

@relation movie_data 

@attribute movie1_one {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie1_two {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie1_three {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie2_one {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie2_two {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute movie2_three {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute decision_one {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute decision_two {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 
@attribute decision_three {28,12,16,35,80,105,99,18,82,2916,10751,10750,14,10753,10769,36,10595,27,10756,10402,22,9648,10754,1115,10749,878,10755,9805,10758,10757,10748,10770,53,10752,37} 

@data 
18,27,53,18,53,10756,18,27,53 
35,18,10749,18,10769,18,18,10769,18 
16,878,53,16,18,16,16,18,16 
35,10749,10757,18,18,18,18,18,18 
80,18,10748,18,10749,18,18,10749,18 
28,18,36,35,18,10751,28,18,36 
18,10749,10769,35,18,10402,35,18,10402 
28,12,878,18,10749,10769,18,10749,10769 
35,10749,35,14,10402,10751,14,10402,10751 

Answer 1

Если я правильно понял вас, у вас возникла проблема «Multi-Class» или «Multi-Target». У вас есть несколько простых вариантов решения проблемы:

1. Создать новый целевой класс, который включает в себя все 3 (конкатенацию decision_one, decision_two и decision_three)

2. поезд каждой цели отдельно.

Answer 2

Я думаю, что самый простой подход был бы, как сказал Белла, чтобы обучать три отдельные модели, по одному для каждого класса, возможно, удаление остальной части класса attribs (в зависимости от того, хотите ли вы другие два класса влиять на вашу классификацию).