Follow up on issue https://github.com/Jack-H-Buckner/UniversalDiffEq.jl/issues/27. Both NUTS! and SGLD! create a vector of parameters, where each entry is a draw of the posterior distribution. All forecasting functions need to account for this, for each parameter combination in the posterior, one can do an individual prediction and calculate the percentiles of the confidence interval using StatsBase.percentile.
jarroyoe
commented
3 weeks ago
Follow up on issue https://github.com/Jack-H-Buckner/UniversalDiffEq.jl/issues/27. Both
NUTS!andSGLD!create a vector of parameters, where each entry is a draw of the posterior distribution. All forecasting functions need to account for this, for each parameter combination in the posterior, one can do an individual prediction and calculate the percentiles of the confidence interval usingStatsBase.percentile.