ArrayOutOfBounds exception with ItemKNN
diarmuidmorgan opened this issue · 2 comments
Hi apologies in advance if I'm just doing something wrong.
I'm running librec itemknn from the commandline with the config listed in the algorithm list.
It will repeatedly crash with the following output.
19/03/25 12:45:41 INFO TextDataConvertor: Dataset: [./librecdata/depaul.csv] 19/03/25 12:45:41 INFO TextDataConvertor: DataSet: ./librecdata/depaul.csv is finished 19/03/25 12:45:42 INFO TextDataConvertor: rating Scale: [1.0, 2.0, 3.0, 4.0, 5.0] 19/03/25 12:45:42 INFO TextDataConvertor: user number: 97, item number is: 79 19/03/25 12:45:42 INFO TextDataModel: Transform data to Convertor successfully! 19/03/25 12:45:42 INFO TextDataModel: Split data to train Set and test Set successfully! 19/03/25 12:45:42 INFO TextDataModel: Data cardinality of training is 1144 19/03/25 12:45:42 INFO TextDataModel: Data cardinality of testing is 299 19/03/25 12:45:43 INFO ItemKNNRecommender: Job Setup completed. 19/03/25 12:45:43 INFO ItemKNNRecommender: Job Train completed. Exception in thread "main" java.lang.ArrayIndexOutOfBoundsException: 79 at net.librec.math.structure.VectorBasedDenseVector.get(VectorBasedDenseVector.java:91) at net.librec.recommender.cf.ItemKNNRecommender.predict(ItemKNNRecommender.java:132) at net.librec.recommender.MatrixRecommender.predict(MatrixRecommender.java:272) at net.librec.recommender.MatrixRecommender.recommendRating(MatrixRecommender.java:238) at net.librec.recommender.MatrixRecommender.recommendRating(MatrixRecommender.java:219) at net.librec.eval.EvalContext.<init>(EvalContext.java:50) at net.librec.eval.EvalContext.<init>(EvalContext.java:57) at net.librec.job.RecommenderJob.executeEvaluator(RecommenderJob.java:213) at net.librec.job.RecommenderJob.executeRecommenderJob(RecommenderJob.java:131) at net.librec.job.RecommenderJob.runJob(RecommenderJob.java:89) at net.librec.tool.driver.RecDriver.run(RecDriver.java:84) at net.librec.tool.driver.RecDriver.main(RecDriver.java:108)
Am I doing something wrong?
Thanks,
Diarmuid
Looking through the source code of cf.ItemKNNRecommender I see this line in the predict method:
if (isRanking) { return predictValue; } else { return predictValue > 0 ? itemMeans.get(userIdx) + predictValue / simSum : globalMean; } }
Else where in the train method, itemMeans is initialized with dimension n_users
`protected void trainModel() throws LibrecException {
itemMeans = new VectorBasedDenseVector(numItems);
double globalMean = trainMatrix.mean();
itemList = new ArrayList<>();
for (int itemIndex = 0; itemIndex < numItems; itemIndex++) {
itemList.add(itemIndex);
}
itemList.parallelStream().forEach(itemIndex -> {
SequentialSparseVector itemRatingVector = trainMatrix.column(itemIndex);
itemMeans.set(itemIndex, itemRatingVector.getNumEntries() > 0 ? itemRatingVector.mean() : globalMean);
});
createItemSimilarityList();
}`
Should it be itemMeans.get(itemIdx) instead?
@diarmuidmorgan I got the same issue, and I agree with you. It should be itemMeans.get(itemIdx)
;
@wangyufengkevin could you fix it?