Modelling multivariate data using HGF

[ddgpalmer]

I'm trying to model responses for an experiment where targets are randomly presented on the circumference of a circle. To convert the circular data of target location to linear data, it is sine and cosine transformed, but then necessarily becomes multivariate data. The documentation for the hgf package, and in particular, the tapas_fitModel() function say that it can accept a matrix as input, but when I input my data as a two column matrix, I get the same result as if only supplying the model with the first column. I also note that in the original paper describing the HGF, it is explicitly stated that the model is applicable to multivariate data (although I realize this doesn't necessariliy guarantee the software can).

Is there a way of getting the tapas HGF model to work on multivariate data that I am missing?

[alexjhess]

Hi @ddgpalmer,

You are correct, the HGF framework is in fact able to handle multimodal data. However, you need to make sure that you specify an appropriate observation model that handles the response data in matrix format appropriately. Check out https://www.biorxiv.org/content/10.1101/2024.02.19.581001v1 for an example application of the HGF to multimodal data (choices + RT in this case). Links to code and data are also included, so you should be able to look at the code for the response models that we used (https://gitlab.ethz.ch/tnu/code/hessetal_spirl_analysis/-/tree/master/comb_obs_models?ref_type=heads). Hope this helps!

Cheers,
Alex