Unpredictable Results Using NaiveBayesGenerator

[sethjuarez]

Unpredictable Results Using NaiveBayesGenerator

[sjwoodard]

Changing the training percentage from 0.80 to 0.75 resolved the discrepancy. I think this is probably the expected result of using the Naive Bayes model, and this issue can be closed. Thanks again!

[sethjuarez]

Awesome. I got worried!

[groffg]

Update: created separate issue for this; please disregard this comment

I must be doing something horribly wrong. I'm trying to use naive bayes to categorize some data based on input via a spreadsheet. The code runs, but it seems to give me the same category regardless of my input. I've appended some sample code below (gloss over the helper call that reads Excel; that just takes the sheet & converts to a .net dataset). Also, I'm limiting the training data to the first 100 rows. Any more than that and the program gets an out of memory exception!

So I have two questions:

• how to get a meaningful prediction/classification?
• how to load much larger training dataset without memory overflow?

    [Serializable]
    public class PeerCategory
    {
        [Feature]
        public string PeerCategoryDesc { get; set; }
        [Label]
        public string CAPeerCategory { get; set; }
    }

    [TestFixture]
    //[Ignore]
    public class BayesTest
    {

        [Test]
        public void TestExcel()
        {
            string pathToFile = @"C:\data\Training Data.xlsx";
            using (var reader = new FileReaderExcel(pathToFile))
            {
                int index = reader.GetSheetIndexByName("V6");
                var dt = reader.GetDataTableFromExcelSheet(index, true);
                var peerCatList = (
                    from DataRow row in dt.Rows
                    select new PeerCategory()
                    {
                        PeerCategoryDesc = row.Field<string>(0), 
                        CAPeerCategory = row.Field<string>(1)
                    }).Take(100).ToList();


                var width = peerCatList.Select(t => t.CAPeerCategory).Distinct().Count();

                IGenerator generator = new NaiveBayesGenerator(width);
                generator.Descriptor = Descriptor.Create<PeerCategory>();
                LearningModel learned = Learner.Learn(peerCatList, 0.8, 1000, generator);

                IModel model = learned.Model;
                double accuracy = learned.Accuracy;


                var value1 = model.Predict(GetItem("paint"));
                var value2 = model.Predict(GetItem("flower"));
                var value3 = model.Predict(GetItem("sofa"));
                var value4 = model.Predict(GetItem("desk"));
                var value5 = model.Predict(GetItem("bones"));
                                // value1 thru 5 will be the same category over and over again!!

            }

        }

        public PeerCategory GetItem(string desc)
        {
            var item = new PeerCategory()
            {
                PeerCategoryDesc = desc,
                CAPeerCategory = string.Empty
            };
            return item;
        }