So I had a basic database that consisted of a single table with a bunch of questions for each column (Q1, Q2, Q3) and a respondant ID for the person answering the question.
So I created a Naive Bayes and Neural Net model and was exploring the results for any interesting tidbits. When I was exploring relationships based on one of 3 Questions I was trying to generate predictions for, I applied a model filter on that question
to get rid of responses if there weren't enough responses and more specifically to eliminate blanks (which indicate that the answer for the question I was interested in was skipped).
Everything was working fine, except I went back to one question Q72, to double check some results - and BAM whenever I apply a filter to eliminate blanks on that question I get NO results for the Naive Bayes Browse (I.e. when I get to Attribute
discrimination there are no values for Value 1 and Value 2, and even more peculiar, nothing shows up under Attribute Profile).
But here is the weird thing, I double check the filter by running a select on the table using the filter as a WHERE, and rows appear (so its not a logic problem). I apply the EXACT same filter on my Neural Net for Q72 and everything is fine.
I've tried reprocessing everything, making a new project with a new mining structure (another Naive Bayes), I've even tried applying a filter at the Data Source level so that I wouldn't have to apply one at the Model level - Same thing every time No Results
for Q72. I can apply filters for the other two questions (Q40 and Q48) on the Naive Bayes, and it works. Just for some reason, I can't apply a filter of any kind for Q72 either at the model or the source level. Also - I am 98%
sure that a mere 7 hours ago, everything was working fine, even for Q72.
Does anyone know what is going on or have any suggestions I can use to get to the bottom of this bizarre and incredibly frustrating behaviour?
Oh - and if I process and Browse the Naive Bayes for Q72 WITHOUT a filter of any kind I get results that are completely different than what I got before, and in fact about 70% of the results in the Attribute Discrimination have a value of "Missing" for
each row where it shows you whether if favors Value 1 or Value 2