Evaluation.java

package org.shanbo.feluca.cf.common;

import org.shanbo.feluca.data2.DataEntry;
import org.shanbo.feluca.data2.RandomAccessData;
import org.shanbo.feluca.data2.Vector;
import org.shanbo.feluca.data2.DataStatistic;
import org.shanbo.feluca.paddle.common.Utilities;

public class Evaluation {

	/**
	 * testing RMSE using history 
	 * @param test
	 * @param model
	 * @return
	 * @throws Exception
	 */
	public static double runRMSE(RandomAccessData train, DataEntry test, Recommender model) throws Exception{
		double error = 0;

		int dbsize = Utilities.getIntFromProperties(test.getDataStatistic(), DataStatistic.TOTAL_FEATURES);
		System.out.println(dbsize);
		int cc = 0;
		test.reOpen();
		for(Vector v = test.getNextVector(); v != null ; v = test.getNextVector()){
			UserRatings ur = new UserRatings(v);

			int[] toPredicts = new int[ur.getItemNum()];
			// draw itemid apart
			for(int i = 0 ; i < ur.getItemNum(); i++){
				toPredicts[i] = ur.getRatingByIndex(i).itemId;
			}
			UserRatings history = new UserRatings(train.getVectorById(ur.getUid()));
			// predict in batch mode
			float[] predicts = model.predict(history, toPredicts);
			if (predicts == null)
				dbsize -= ur.getItemNum();
			else{
				for(int i = 0 ; i < ur.getItemNum(); i++ ){
					RatingInfo ri = ur.getRatingByIndex(i);
					error += Math.pow((ri.rating - predicts[i]), 2);
				}
			}
			cc += 1;
			if (cc % 2000 == 0){
				System.out.print(".");
			}

		}
		test.close();
		train.close();
		return Math.sqrt(error / dbsize);
	}

	/**
	 * without cf history
	 * @param test
	 * @param model
	 * @return
	 * @throws Exception
	 */
	public static double runRMSE(DataEntry test, Recommender model) throws Exception{
		double error = 0;

		int dbsize = Utilities.getIntFromProperties(test.getDataStatistic(), DataStatistic.TOTAL_FEATURES);
		System.out.println(dbsize);
		int cc = 0;
		test.reOpen();
		for(Vector v = test.getNextVector(); v != null ; v = test.getNextVector()){
			UserRatings ur = new UserRatings(v);

			int[] toPredicts = new int[ur.getItemNum()];
			// draw itemid apart
			for(int i = 0 ; i < ur.getItemNum(); i++){
				toPredicts[i] = ur.getRatingByIndex(i).itemId;
			}
			// predict in batch mode
			float[] predicts = model.predict(ur, toPredicts);
			if (predicts == null)
				dbsize -= ur.getItemNum();
			else{
				for(int i = 0 ; i < ur.getItemNum(); i++ ){
					RatingInfo ri = ur.getRatingByIndex(i);
					error += Math.pow((ri.rating - predicts[i]), 2);
				}
			}
			cc += 1;
			if (cc % 2000 == 0){
				System.out.print(".");
			}
		}
		test.close();
		return Math.sqrt(error / dbsize);
	}

}