pre-configured ModelChains

Is your feature request related to a problem? Please describe. ModelChain has a lot of options and can be challenging to configure. This also makes it difficult to implement reference implementations of workflows.

Describe the solution you’d like Create wrappers for pre-configured ModelChains. For example, a PVWatts version of ModelChain might look like:

# modelchain.py
PVWatts = functools.partial(
    ModelChain, 
    dc_model='pvwatts', ac_model='pvwatts', losses_model='pvwatts', 
    transposition_model='perez', aoi_model='physical'
)

# user code
my_model_chain = PVWatts(my_location, my_system)

This SO post discusses some subtleties with using partial with classes, so the actual implementation might require the helper function in that post.

Describe alternatives you’ve considered We could subclass ModelChain but that seems like overkill for this. Inheritance leads to brittle code, in my experience, so I prefer to avoid it if possible.

We could supply dicts of parameters e.g.

# modelchain.py
pvwatts_config = dict(
    dc_model='pvwatts', ac_model='pvwatts', losses_model='pvwatts', 
    transposition_model='perez', aoi_model='physical'
)

# user code
my_model_chain = ModelChain(my_location, my_system, **pvwatts_config)

Any other ideas?

[wholmgren]

Factory pattern is a good idea. Location has a couple of factories like from_tmy, so it fits with the library. I think I’d go for a class method on the ModelChain class but don’t feel strongly about it.

We might want a different method for each predefined chain because they won’t necessarily have the same parameters. For example, a PVWatts chain would not expose transposition to the user (documentation says to use Perez) but another chain might leave that to the user as a kwarg.

I could see similar factories for PVSytem that help people avoid the less-than-explicit API of “put these keys in modeling_parameters and these keys in inverter_parameters”.

(I fixed the link in @kanderso-nrel’s post)

[cwhanse]

In #1022, I like having the PVWATTS_CONFIG dict for maintenance. I prefer the ModelChain.with_pvwatts method over the subclass, but not strongly. I think the with_pvwatts style is more explicit and easier to understand than the subclass approach.