Conditional fits with fixed non-POI parameters

[balunas]

I've got a question which feels like it should very simple, but I can't seem to find any indication of how to do it in the documentation. I've got a likelihood model with a pretty common structure: some yields in bins from different sources, which each have a few modifiers attached. These represent systematic NPs or unconstrained normalization parameters in the fit, etc. I want to run a simple MLE fit holding some of these parameters fixed at values that I specify at the python level (and allowing all the others to vary as usual). I know that there's a dedicated function that does this holding the POI fixed, but I want to fix other parameters. I'd be surprised if the python API doesn't support this, but I can't seem to find it documented anywhere. Does anyone know how to do this?

[alexander-held]

The infer.mle.fit API takes a list of initial parameter values and a list of booleans to denote whether a given parameter is meant to be held constant. If it is set to constant, it will take the initial parameter value. So you can use model.config.par_names to identify the position in the list of parameters that a given parameter you want to hold constant is located at, then use model.config.suggested_init() and model.config.suggested_fixed() to get the default settings for initial values / fixed parameters, edit these lists as desired (parameter order matches model.config.par_names), and pass the new lists to infer.mle.fit while using the init_pars and fixed_params keyword arguments.

[balunas]

Perfect, thanks - it was there in the API all along, I apparently just missed it or was looking in the wrong place. Sorry for the silly question!