Closed stephengreen closed 1 year ago
Nice work, I think it looks reallly good overall! Getting all those corner plots to work must have been hard!
I don't think there is a KDE option for the marginals in corner, but at some point when I used corner I wrote a script to automatically do this. But chainconsumer has this by default. That being said I don't know if it's worth it to do at the moment since we want to start testing on conda-forge ASAP.
I actually asked GPT-4 how to fully smooth the 1D marginals, and it wrote a code to plot those axes with seaborn. This seemed excessive; I think we can live with the less smooth ones.
If GPT-4 can't do it there's no chance for us :)
I think what I was doing was individually taking the 1d and 2d marginals and training KDEs on them, and then resampling the KDEs. This in general does take a long time to create the plot. But to be honest, I agree the less smooth ones are fine. Maybe something to implement down the line if there is time.
Ah also, should we be returnning the figure object? I think it could be quite useful if the user wants to edit the figure or add things on later?
Ah also, should we be returnning the figure object? I think it could be quite useful if the user wants to edit the figure or add things on later?
Fine with me. In general I wasn't trying to leave a huge amount of flexibility to the user, but if you think this is useful please go ahead.
Ah right, I mean to change that back to how it was before. I can fix it when I'm back to my laptop.
This replaces chainconsumer with corner for making corner plot, as described in #166. It places a multiple-posterior corner plotting function in
dingo.core.utils.plotting
.I have tested it in
dingo_pipe
, but not ininference_pipeline.py
anddiagnostics.py
. Could someone who uses these please test it?