Inconsistent HDI in posterior plot and table summary

[chiukenny]

The HDI that is in summaries does not match the red line on the posterior plot for the BasicReparameterized model in this tutorial. For μ=0.5, the summary reports an upper HDI of 0.452, but the plot shows an upper value >0.5.

To reproduce: run Exercise 1 in the tutorial.

[alexandrebouchard]

@kevinchern sorry to add another thing to you plate, but do you mind looking at this when you have a chance? Thank you!

[kevinchern]

:O.... on it!

[alexandrebouchard]

Thanks!

…

On Wed, Jan 27, 2021, 11:35 Kevin, ***@***.***> wrote: :O.... on it! — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#163 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAZKDKL4BKJE3SWMCGIICUDS4BTIFANCNFSM4WVZKIGQ> .

[kevinchern]

Looks like the current plotting template automatically includes a burn-in, whereas summaries do not.
I think we should make burn-ins consistent across outputs, or label the plots to indicate burn-in has been applied.
Also worth considering setting the default to 0 burn-in instead of 0.5.

@alexandrebouchard any suggestions?

[alexandrebouchard]

Most plot actually discard burn in right now. For example, in the posterior plotting code, the line

blangSDK/src/main/java/blang/runtime/internals/DefaultPostProcessor.xtend

Line 434 in 6b4b47f

«removeBurnIn»

triggers the following which removes the burn in samples:

blangSDK/src/main/java/blang/runtime/internals/DefaultPostProcessor.xtend

Line 518 in 6b4b47f

def String removeBurnIn() {

Ah I see, it's actually the other way around: summaries don't remove the burn in right now:

blangSDK/src/main/java/blang/runtime/internals/DefaultPostProcessor.xtend

Line 589 in 6b4b47f

val directory = outputFolder(Output::summaries)

I think adding the burn in macro should do it. Do you mind giving it a try? Thanks again!

[kevinchern]

Yep! Discarding burn-in from summaries gives consistent results.

[alexandrebouchard]

Resolved in #164