OpenCV中的决策树

Question

I have a table with information about the weather and the number of forest fires in several days in a region:

 Temperature | Humidity | Wind | ... | Number of forest fires

All data are presented as numbers.

It is necessary to use these data to build a decision tree, and using it and input parameters (temperature, humidity, wind speed, etc.) to predict the number of forest fires.

For working with decision trees in OpenCV there is a class CvDTree . To construct a decision tree in this class has a method of train :

boolean train (Mat trainData, int tflag, Mat responses)

How to convert my input (temperature, humidity, wind, ..., number of forest fires) in Mat trainData ?

Added:

I get an error during compiling this code:

package com.selw;
import org.opencv.core.Core;
import org.opencv.core.CvType;
import org.opencv.core.Mat;
import org.opencv.core.Scalar;
import org.opencv.ml.DTrees;
import org.opencv.ml.Ml;

public class Main {

static{ System.loadLibrary(Core.NATIVE_LIBRARY_NAME); }

public static void main(String[] args) {
    Mat data = new Mat(5, 3, CvType.CV_32FC1, new Scalar(0));
    data.put(0, 0, new float[]{1.69f, 1, 0});
    data.put(1, 0, new float[]{1.76f, 0, 0});
    data.put(2, 0, new float[]{1.80f, 0, 0});
    data.put(3, 0, new float[]{1.77f, 0, 0});
    data.put(4, 0, new float[]{1.83f, 0, 1});

    Mat responses = new Mat(5, 1, CvType.CV_32SC1, new Scalar(0));
    responses.put(0, 0, new int[]{0, 1, 1, 0, 1});

    DTrees tree = DTrees.create();
    tree.train(data, Ml.ROW_SAMPLE, responses);
    }
}

Error (in line tree.train(data, Ml.ROW_SAMPLE, responses); ):

Exception in thread "main" java.lang.Exception: std::exception: vector too long at org.opencv.ml.StatModel.train_0(Native Method) at org.opencv.ml.StatModel.train(StatModel.java:95) at com.selw.Main.main(Main.java:35) at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method) at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62) at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43) at java.lang.reflect.Method.invoke(Method.java:497) at com.intellij.rt.execution.application.AppMain.main(AppMain.java:140)

Answer 1

The Mat data object contains the sample data that the tree will train on. One row corresponds to one sample. The more samples you have, the better trained your tree will be. You can have as many columns as you want in your Mat data object. Each column corresponds to a different sample feature.

The Mat responses object contains values that the tree aims to predict. This is a one column matrix with the same number of rows as your Mat data object. Each row number in your Mat responses object corresponds to the sample with the same row number in your Mat data object.

For your example of forest fires being predicted based on a region's measurements of Temperature, Humidity, and Wind... each of the region's measurements need to be converted into floating point values. Each region will have one row in your Mat data object, and one row in your Data responses object. The Mat data row will be filled with the region's measurements expressed as floating point values. The Mat responses row will be the number of forest fires (or as Nicholas pointed out in your comments, it might be better to place the number of forest fires into indexed histograms).

The current error in your code can be fixed by changing these settings:

    tree.setCVFolds(1); // opencv has not implemented pruning with cross-validation yet
    tree.setMaxDepth(10); // there is currently a bug that will run the depth of the tree to infinity if you do not set a max

I was able to compile and run without errors after changing those.