Wrong log_pdf evaluations returned?

[MichaelClerx]

@MarcBerliner wrote:

Calculating my objective function is expensive, so I’d like to store the log_pdf values alongside the samples to avoid recalculating them.

As a side note, I noticed that samplers that successfully returned a log_pdf vector returned values that differed from my objective function. I suspect the samplers are applying some internal rescaling, possibly related to the Jacobian or covariance matrix. It would be great if this rescaling could be removed from the log_pdf values.

Originally posted in #1691 (comment)

[MichaelClerx]

@MarcBerliner do you happen to have an example handy? Can't think of any log_pdf transformations off the top of my head so bit surprised?

[MichaelClerx]

One thing that does happen, is that it will store a (LogPDF (posterior), LogLikelihood, and Prior) if priors are used. Is there a chance that this is happening and you're looking at the wrong one?