Model for task with missing data and response deadline?

[chutcher]

Hi,

I have a task that I am working to model with a DDM, but for which decisions had to be made within 3 seconds, leading to a fair number of missing responses. I've been looking at the HSSM documentation and see that in a previous release (v 0.2.1) the ability to model missing responses and deadlines was added. However, I am not 100% clear on how to actually take advantage of this functionality. Based on this discussion (#173) I gather that I might need to add a deadline column to my data frame, presumably specifying a 3 second limit, and perhaps use one of the onnx files, along with a missing_data=True flag in my model and the path to the onnx file provided as a likelihood function. But I am not sure which onnx file might be most appropriate, and am a little uncertain whether my other assumptions are correct. Apologies if I've missed the documentation or further discussion already posted somewhere, but I'd love some advice!

Thanks in advance,
Cendri

[AlexanderFengler]

Hi @chutcher,

We front-loaded a bit of the HSSM infrastructure on this, but let it lay dormant while focusing on some other issues in the meanwhile, e.g. the documentation part.

The idea is that you can provide data with missing (coded) values and use a [model]_deadline model string and the backend will then adjust the likelihood computations accordingly.

What happens internally is that the missing values are going to be assigned to a network that computes the probability of omission given the parameters and deadline for a given trial, so this will in principle work for any of the models.

We can prioritize adding more docs on this functionality inc. a tutorial within a week or two if that helps.

Best,
Alex