Different OFV in Pumas compared to NONMEM

Diagnostics
1

Hi, I was wondering if Pumas and NONMEM calculate OFV in different ways? A model I converted from NONMEM to pumas is identical in every way besides the OFV

2

hi Pawan -

Welcome to the community!

Pumas reports the loglikelihood whereas NONMEM reports the -2 x loglikelihood.

In Pumas you can do metrics_table(model_fit) where you get a table of metrics which should contain the -2LL that you can compare with NONMEM.

Hope this helps

Vijay

3

NONMEM also leaves out the normalization constant of the Gaussian likelihood in their OFV so the exact relationship between the two is

NONMEM_OFV = -2*loglikelihood(_fitted_model) - Pumas.nobs(_fitted_model)*log(2π)

Notice that this only holds true for purely Gaussian models and not, e.g. truncated (M2) or censored (M3) models.

2 Likes
4

Thank you, @andreasnoack and @vijay. Using the above relationship, I am able to get the same NONMEM OFV .