Domain Error Message in version 2.6.1

Estimation
1

Hi there,

Could you please help me understand why I have the error below in version 2.6.1 when I run the ‘fit()’ argument?

[ Info: Checking the initial parameter values.
ERROR: DomainError with Inf:
The initial parameter values cause a negative log likelihood of Inf. This can be due to model misspecification or bad initial parameter values.
Stacktrace:
 [1] _optim_check_initial_values(costf::Pumas.var"#611#615"{…}, vparam::Vector{…}, verbose::Bool)
   @ Pumas C:\build\run\_work\PumasSystemImages\PumasSystemImages\julia_depot\packages\Pumas\GJfPM\src\estimation\likelihoods.jl:4450
 [2] _fit(m::PumasModel{…}, population::Vector{…}, param::@NamedTuple{…}, approx::LaplaceI{…}, ensemblealg::EnsembleThreads, optimize_fn::Pumas.DefaultOptimizeFN{…}, fixedparamset::ParamSet{…}, fixedparam::@NamedTuple{…}, checkidentification::Bool, diffeq_options::@NamedTuple{}, init_vrandeffsorth::Vector{…}, verbose::Bool, ignore_numerical_error::Bool, optim_state::Nothing)
   @ Pumas C:\build\run\_work\PumasSystemImages\PumasSystemImages\julia_depot\packages\Pumas\GJfPM\src\estimation\likelihoods.jl:4622
 [3] __fit
   @ C:\build\run\_work\PumasSystemImages\PumasSystemImages\julia_depot\packages\Pumas\GJfPM\src\estimation\likelihoods.jl:4331 [inlined]
 [4] __fit
   @ C:\build\run\_work\PumasSystemImages\PumasSystemImages\julia_depot\packages\Pumas\GJfPM\src\estimation\likelihoods.jl:4318 [inlined]
 [5] fit(m::PumasModel{…}, population::Vector{…}, param::@NamedTuple{…}, approx::LaplaceI{…}; optim_alg::Nothing, optim_options::@NamedTuple{…}, optimize_fn::Nothing, constantcoef::Tuple{}, omegas::Tuple{}, ensemblealg::EnsembleThreads, checkidentification::Bool, diffeq_options::@NamedTuple{}, init_randeffs::Nothing, init_vrandeffsorth::Nothing, verbose::Bool, ignore_numerical_error::Bool)
   @ Pumas C:\build\run\_work\PumasSystemImages\PumasSystemImages\julia_depot\packages\Pumas\GJfPM\src\estimation\likelihoods.jl:4285
 [6] top-level scope
   @ d:\OneDrive - Nexus365\TB_Pumas\__tte_julia_pumas_nonmem\__tte_isoniazid\__tte_isonizaid_pumas\__pk_models_isoniazid_finals.jl:215
Some type information was truncated. Use `show(err)` to see complete types.

The same model (below) ran without error in version 2.5, and also in version 2.6.1 when ‘censoring’ was removed.

image
image551×827 41.7 KB

Thanks.

2

I think the issue might be that you have data points that are below the lower censoring point. Previously, we interpreted any points below the censoring point to belong to the censored category but with Pumas 2.6.0, such data points are assigned zero probability which is the same a -Inf log-likelihood. Hence, please check if you have any data points that are smaller than log(12). Please also change censored(...) to censored_latent(...). It shouldn’t matter for estimation but it will make a difference when predicting as it will then predict the uncensored variable instead of the censored variable.

3

Hi Andreas,

Yes, you’re right. I set values <BQL to LLOQ/2 in the dataset. How do you then suggest I handle these values in the dataset? Set them to ‘0’?

Warm regards