问题
I am generating decision trees in Weka in Java code as follows:
J48 j48DecisionTree = new J48();
Instances data = null;
data = new Instances(new BufferedReader(new FileReader(dt.getArffFile())));
data.setClassIndex(data.numAttributes() - 1);
j48DecisionTree.buildClassifier(data);
Can I save the results of the Weka results buffer to a text file in the program, such that the following can be saved at run-time to a text file:
=== Stratified cross-validation === === Summary ===
Correctly Classified Instances 229 40.1754 %
Incorrectly Classified Instances 341 59.8246 %
Kappa statistic 0.2022
Mean absolute error 0.1916
Root mean squared error 0.3138
Relative absolute error 80.8346 %
Root relative squared error 91.1615 %
Coverage of cases (0.95 level) 96.3158 %
Mean rel. region size (0.95 level) 70.9774 %
Total Number of Instances 570
=== Detailed Accuracy By Class ===
TP Rate FP Rate Precision Recall F-Measure ROC Area Class
0.44 0.012 0.786 0.44 0.564 0.76 Business and finance and economics
0 0 0 0 0 0.616 Fashion and celebrity lifestyle
0.125 0.01 0.667 0.125 0.211 0.663 Film
0 0.002 0 0 0 0.617 Music
0.931 0.78 0.318 0.931 0.474 0.633 News and current affairs
0.11 0.006 0.786 0.11 0.193 0.653 Science and nature and technology
0.74 0.012 0.86 0.74 0.796 0.85 Sport
Weighted Avg. 0.402 0.224 0.465 0.402 0.316 0.667
=== Confusion Matrix ===
a b c d e f g <-- classified as
22 0 0 0 25 2 1 | a = Business and finance and economics
0 0 1 0 59 0 0 | b = Fashion and celebrity lifestyle
0 0 10 1 69 0 0 | c = Film
0 0 1 0 69 0 0 | d = Music
5 0 2 0 149 0 4 | e = News and current affairs
1 0 0 0 87 11 1 | f = Science and nature and technology
0 0 1 0 11 1 37 | g = Sport
dt is an instance of a class of mine to represent decision tree details.
As I'm running a large number of classifiers, this would help somewhat.
回答1:
The Weka classifiers have an extensive #toString() method, which gives you a human readable representation, in this case the tree. You can also use #toSource(String) to get a Java code equivalent for the decision tree.
If you want to store the model for re-using it later, have a look at weka.core.SerializationHelper.
回答2:
Yes, this can be done. But you need to create an instance of Evaluation in Weka and call the appropriate methods from the instance:
Evaluation eval = new Evaluation(data);
eval.evaluateModel(j48DecisionTree, data);
System.out.println(eval.toSummaryString("\nResults\n======\n", true));
Will give a summary.
But then methods such as:
eval.pctCorrect();
Can be called. See Weka Javadoc for further info.
来源：https://stackoverflow.com/questions/12699059/writing-the-results-of-weka-classifier-to-file-in-java