Predictive Analytics-Naive Bayes-Error

[vujennyfer]

I am trying to use the Naive Bayes Classifier tool to determine if a review is "negative" or "positive". I am able to produce probability results for the training dataset, but not for the test set (unseen data). I keep getting the below errors. Attached is the workflow and the dummy data. The excel sheet has two tabs for the training and test data set. Any help is appreciated!! 

[apathetichell]

Doesn't the Alteryx Naive-Bayes tool require two predictors? I know you above version seems to work - but this could be triggering the error.

More evidence - from the tool sample workflow...

The Naive Bayes Classifier tool assumes that all predictors are independent of one another and predicts, based on a sample input, a probability distribution over a set of classes.

To configure the tool, first name the model. Next, select the target and at least 2 predictors. Choose a positive value as a smoothing parameter. The default value is 0. This feature allows you to smooth the data by accounting for class/feature combinations that might either be entirely absent from the training set, or are otherwise under-represented in frequency.

[vujennyfer]

Hi @apathetichell, thanks for your reply. I added another predictor (ratings) to have two total predictors but I am still seeing those errors. I also changed the Laplace smoothing to '1' as a positive value. 

[apathetichell · Accepted solution]

made a few changes and this seems to work.

I think the test data and train data need to have the same fields - so the "positive/negative" was set to a dummy field.  Made a few other small changes here and there...

I'm not ecstatic with this version so if no one has given you a better answer (as to why this works and some changes don't) - I'll try to get back to this in a bit.

[vujennyfer]

THANK YOU SOO MUCH @apathetichell!! This worked perfectly. You're the BEST! 🙂 

[vujennyfer]

Hi @apathetichell, I'm returning to this problem and wondering if there is a way (maybe not Naive Bayes but another tool) that could help with text classification in Alteryx. For this solution to work, we needed at least 2 fields (review & rating). Is it possible to only utilize one field (review) to truly get the model to learn text classification instead of using the numerical rating? 

[apathetichell · Accepted solution]

Hi! I don't know about the text mining processes in Alteryx (extra $) but perhaps someone can talk to you about them? I know R has various NLP/stemming/document mapping functions which you could add via the R tool (for free) as does Python (for free). Python is python so if you start a new thread as "implementing text classification in python" someone will probably respond.

[vujennyfer]

Thanks for your input @apathetichell! I will try those suggestions 🙂