All Downloads are FREE. Search and download functionalities are using the official Maven repository.

com.actelion.research.calc.regression.ModelError Maven / Gradle / Ivy

There is a newer version: 2024.11.2
Show newest version
/*
 * Copyright (c) 1997 - 2016
 * Actelion Pharmaceuticals Ltd.
 * Gewerbestrasse 16
 * CH-4123 Allschwil, Switzerland
 *
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 * 1. Redistributions of source code must retain the above copyright notice, this
 *    list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright notice,
 *    this list of conditions and the following disclaimer in the documentation
 *    and/or other materials provided with the distribution.
 * 3. Neither the name of the the copyright holder nor the
 *    names of its contributors may be used to endorse or promote products
 *    derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

package com.actelion.research.calc.regression;

import com.actelion.research.calc.Matrix;
import com.actelion.research.calc.MatrixFunctions;
import com.actelion.research.calc.classification.PrecisionAndRecall;
import com.actelion.research.util.Formatter;
import com.actelion.research.util.datamodel.DoubleArray;

import java.util.*;

/**
 * ModelError
 *
 * This class is a data model for the error. It is not the error of the model.
 *
 * @author Modest von Korff
 * Aug 14, 2015 MvK Start implementation
 */
public class ModelError {


	// Average from the sum of |errors|
	public double error;

	public double errorMedian;

	public double errorRelative;

	public double errorRelativeMedian;

	public double errorRelativeWeighted;

	public double errSumSquared;

	public double errMax;

	public double errMin;
	
	public double corrSquared;
	public double corrSquaredSpearman;

	// Classification
	public boolean classification;

	public PrecisionAndRecall precisionAndRecall;

	public boolean failed;

	public int nNotFiniteRelError;

	/**
	 * 
	 */
	public ModelError() {
		failed = false;
	}

	public void setFailed() {
		this.failed = true;
	}

	public boolean isFailed() {
		return failed;
	}

	/* (non-Javadoc)
	 * @see java.lang.Object#toString()
	 */
	@Override
	public String toString() {


		StringBuilder sb = new StringBuilder();

		if(failed) {
			sb.append("ModelError [error=failed]");
		} else {
			sb.append("ModelError [error=");
			sb.append(Formatter.format3(error));
			sb.append(", errRelativeMedian=");
			sb.append(Formatter.format3(errorRelativeMedian));
			if(nNotFiniteRelError!=0){
				sb.append(", NotFiniteRelError=");
				sb.append(nNotFiniteRelError);
			}
			sb.append(", errMax=");
			sb.append(Formatter.format3(errMax));
			sb.append(", errMin=");
			sb.append(Formatter.format3(errMin));
			sb.append(", corrSquared=");
			sb.append(Formatter.format3(corrSquared));

			if(precisionAndRecall!=null){
				sb.append(", Cohen's kappa=");
				sb.append(Formatter.format3(precisionAndRecall.calculateCohensKappa()));
			}

			sb.append("]");
		}
		return sb.toString();
	}

	/**
	 * Calculates the absolute and the relative error.
	 * @param Y
	 * @param YHat
	 * @return
	 */
	public static ModelError calculateError(Matrix Y, Matrix YHat){
		
    	ModelError modelError = new ModelError();

    	modelError.errMax = 0;

    	modelError.errMin = Integer.MAX_VALUE;

		DoubleArray daError = new DoubleArray(Y.rows()*YHat.cols());


		double sumSquared = 0;
		for (int i = 0; i < YHat.cols(); i++) {
        	
            for (int j = 0; j < YHat.rows(); j++) {
            	
            	double e = Math.abs(Y.get(j, i) - YHat.get(j, i));

				sumSquared += e*e;

            	modelError.errMax = Math.max(modelError.errMax, e);
            	
            	modelError.errMin = Math.min(modelError.errMin, e);
            	
            	modelError.error += e;

				daError.add(e);
    		}
		}

        modelError.error = modelError.error / (YHat.rows()*YHat.cols());

		modelError.errSumSquared = sumSquared;

		modelError.errorMedian = daError.median();

		DoubleArray daErrorRelative = new DoubleArray(YHat.rows()*YHat.cols());

		modelError.nNotFiniteRelError=0;
		for (int i = 0; i < YHat.cols(); i++) {

			for (int j = 0; j < YHat.rows(); j++) {

				double y = Y.get(j, i);
				double yHat = YHat.get(j, i);

				double er = getRelativeError(y, yHat);

				if(Double.isFinite(er)){
					daErrorRelative.add(er);
				}else {
					modelError.nNotFiniteRelError++;
				}
			}
		}
		if(daErrorRelative.size()>0) {
			modelError.errorRelative = daErrorRelative.avr();
			modelError.errorRelativeMedian = daErrorRelative.median();
		}

		//
		// Weighted error
		//
		DoubleArray daErrorRelativeWeighted = new DoubleArray(YHat.rows()*YHat.cols());

		for (int i = 0; i < YHat.cols(); i++) {

			for (int j = 0; j < YHat.rows(); j++) {

				double y = Y.get(j, i);
				double yHat = YHat.get(j, i);

				double w = Math.log10(10+y);

				if(Math.abs(y) > Matrix.TINY04) {

					double er = Math.abs((yHat - y) / y) * (1.0/w);

					if(Double.isFinite(er)) {
						daErrorRelativeWeighted.add(er);
					}
				} else {

					double er = Math.abs((yHat - y) / Matrix.TINY04) * (1.0/w);

					if(Double.isFinite(er)) {
						daErrorRelativeWeighted.add(er);
					}
				}
			}
		}

		modelError.errorRelativeWeighted = daErrorRelativeWeighted.avr();

		double corr = 0;
		double corrSpearman = 0;
		try {

			corr = MatrixFunctions.getCorrPearson(YHat, Y);

			corrSpearman = MatrixFunctions.getCorrSpearman(YHat, Y);

		} catch (Exception e) {
			e.printStackTrace();
			System.err.println("YHat");
			System.err.println(YHat.toString());
			System.err.println("Y");
			System.err.println(Y.toString());
		}

		if(!Double.isFinite(corr)){
        	corr=0;
		}

		if(!Double.isFinite(corrSpearman)){
			corrSpearman=0;
		}

        modelError.corrSquared = corr*corr;
        modelError.corrSquaredSpearman = corrSpearman*corrSpearman;

		return modelError;

	}

	public static double getRelativeError(double y, double yHat){
		double er = 0;

		if(Math.abs(y) > Matrix.TINY04) {
			er = Math.abs((yHat - y) / y);

		} else {
			er = Math.abs((yHat - y) / Matrix.TINY04);
		}

		return er;
	}

	public static ModelError calculateError(Matrix Y, Matrix YHat, double threshold, boolean above){
		ModelError me = calculateError(Y, YHat);

		me.precisionAndRecall = new PrecisionAndRecall();

		for (int i = 0; i < YHat.cols(); i++) {

			for (int j = 0; j < YHat.rows(); j++) {

				double y = Y.get(j, i);
				double yHat = YHat.get(j, i);

				if(above) {
					if(y>=threshold && yHat>=threshold) {
						me.precisionAndRecall.truePositive++;
					} else if(y < threshold && yHat < threshold) {
						me.precisionAndRecall.trueNegative++;
					} else if (yHat>=threshold){
						me.precisionAndRecall.falsePositive++;
					} else if (yHat threshold && yHat > threshold) {
						me.precisionAndRecall.trueNegative++;
					} else if (yHat<=threshold){
						me.precisionAndRecall.falsePositive++;
					} else if (yHat>threshold){
						me.precisionAndRecall.falseNegative++;
					}
				}
			}
		}

		me.classification = true;

		return me;
	}



		public static List getError(List liME){
		
		List li = new ArrayList();
		
		for (ModelError modelError : liME) {
			li.add(modelError.error);
		}
		
		
		return li;
	}

	public static ModelError getErrorAverage(List liME){

		ModelError modelErrorAvr = new ModelError();

		for (ModelError modelError : liME) {
			modelErrorAvr.errMax += modelError.errMax;
			modelErrorAvr.errMin += modelError.errMin;
			modelErrorAvr.error += modelError.error;



			modelErrorAvr.corrSquared += modelError.corrSquared;
		}

		int n = liME.size();

		modelErrorAvr.errMax /= n;
		modelErrorAvr.errMin /= n;
		modelErrorAvr.error /= n;
		modelErrorAvr.corrSquared /= n;

		return modelErrorAvr;
	}



	public static Comparator getComparatorError(){
		
		return new Comparator() {
			
			@Override
			public int compare(ModelError o1, ModelError o2) {
				int cmp = 0;
				
				if(o1.error > o2.error){
					cmp=1;
				}else if(o1.error < o2.error){
					cmp=-1;
				}
							
				return cmp;
			}
		};
	}

	public static void main(String[] args) {

		int n = 11;

		double fracNoise = 0.1;

		Random random = new Random();

		double [] a = new double[n];
		double [] b = new double[n];

		for (int i = 0; i < n; i++) {
			a[i] = random.nextDouble();
			b[i] = random.nextDouble();
		}

		ModelError meRaw = ModelError.calculateError(new Matrix(false, a), new Matrix(false, b));

		System.out.println(meRaw.toString());

		Arrays.sort(a);
		Arrays.sort(b);

		ModelError meSort = ModelError.calculateError(new Matrix(false, a), new Matrix(false, b));

		System.out.println(meSort.toString());

		for (int i = 0; i < n; i++) {

			if(random.nextDouble()




© 2015 - 2024 Weber Informatics LLC | Privacy Policy