Data Validation for an Equation-Based Model trained with Continuous Data

The front end.
Post Reply
njharn
Posts: 1
Joined: Thu Jun 18, 2020 4:17 pm

Data Validation for an Equation-Based Model trained with Continuous Data

Post by njharn » Mon Jun 22, 2020 7:02 pm

Hello! I am new to this forum, and I have been working with GeNIe for about a month. I've read through most of the manual, with a focus on parameter learning and data validation. I'm currently trying to cross-validate an equation-based model trained with a continuous dataset.

According to the manual, this would require discretizing both the model and the dataset, matching the bins for both, and then performing validation. However, I'd like to perform cross-validation without the extra step of manually defining the bins and discretization boundaries. Is there a way to accomplish this in GeNIe, and if so, how? Thank you for your help!

-njharn

marek [BayesFusion]
Site Admin
Posts: 283
Joined: Tue Dec 11, 2007 4:24 pm

Re: Data Validation for an Equation-Based Model trained with Continuous Data

Post by marek [BayesFusion] » Mon Jun 22, 2020 9:19 pm

Hi Njharn,

I'm afraid you can't do it in GeNIe quite yet but do stay tuned. I have worked recently on a similar problem and even though I had to discretize the model variables for the purpose of structural learning, I wanted to make continuous predictions. I generated an output file in Validation and then converted (in Excel) the probabilities over the discrete class variables into a continuous prediction (took pretty much the middles of the intervals and weighted them by their probabilities). At that point, I was able to calculate the error in prediction. Please do share with others what you came up with -- it is an interesting problem.
Cheers,

Marek

Post Reply