LMInferer should be instantiated with a model (or model path) as a parameter

[gacou54]

Hi!

Users must download the model when instantiating the LMInferer (see below). In some environments (e.g. some Slurm clusters), internet access is restricted, which makes lungmask raise an error.

class LMInferer:
    def __init__(...):
        ...
        # here (mask.py:115), raise an error in restricted network environments
        self.model = get_model(self.modelname)

I suggest that the LMInferer class should have a model parameter instead of a modelname, or at least a modelpath parameter.

Personal opinion: In my mind, LMInferer should not have the responsibility to load or download the model. I prefer the option where the model is given as a parameter

Thank you for this project :)

[JoHof]

Hi,

thanks for your feedback, that's actually a really good point. I think though it should be the responsibility of LMInferer to load the model. Why would you prefer to pass the model directly if may ask? I needs extra code to load the model and pass it to the constructor. I would add just a parameter for modelpath as alternative to the modelname. What do you think?

[gacou54]

Why would you prefer to pass the model directly if may ask? In my mind, the LMInferer role should be to "infer/apply" the model (from a software architecture/engineering point of view). But its only details and it would indeed require you to write more code.

I would add just a parameter for modelpath as an alternative to the modelname. What do you think? It is indeed the simplest fix and I am totally good with it.

Thank you again!

[JoHof]

Hi Gabriel,

I addressed the issue in this PR. Could you please have a look and confirm that this works for you?