Closed bnord closed 5 months ago
This could probably be done with the same code as the Posterior predictive check. The priorpredictive check is just taking in a different set of parameters.
The idea is to be able to make an Nrow x Ncol grid of examples of the training data.
We'd like to be able to order the rows or columns by 1-2 parameters
here's an example with labels on the images of the parameter associated with that postage stamp. images_w_thetaE.pdf
Here's an example of what I'm thinking of (reminder that I only have a linear example)
I'm starting to get used to the 1-d models. I don't know why it's harder to get into that ... maybe it's because these look like 1-1/parity plots. I'm trying to expand my brain on this.
How hard would it be to implement examples based off deepbench to give 2d images as "predicted" outputs from a given parameter?
How hard would it be to implement examples based off deepbench to give 2d images as "predicted" outputs from a given parameter?
I don't think it would be too difficult, I may have to write a test data class but I don't think it'll be too bad. It'll be good to have verification this all works in 2D as well. I'll make an issue for it.
This is the same as the posterior predictive check, except with the priors.
Got the idea from Humna Awan.