The master branch of the code is currently missing any form of error estimation on the marginal statistics. However, there is an implementation of a crude estimate on the error_analysis branch that needs to be merged.
The error estimate uses both samples generated with the density estimate and the original samples to give some indication of the error in the log-probability that can then be propagated through to the marginal stats.
htjb
commented
2 years ago
The master branch of the code is currently missing any form of error estimation on the marginal statistics. However, there is an implementation of a crude estimate on the error_analysis branch that needs to be merged.
The error estimate uses both samples generated with the density estimate and the original samples to give some indication of the error in the log-probability that can then be propagated through to the marginal stats.