How to index a pydantic model using a string object, returning a pydantic model?

[thomasaarholt]

Hi! Thanks for creating a really cool library! It's keeping us sane when reading in config files.

On that topic, we've written the following config file, read in as a dict:

# Minimum working example
{'default_learner': 'model1',
 'learners': {'model1': {'param': {'iterations': 20, 'size': 20}},
              'model2': {'param': {'iterations': 40, 'size': 100}}}}

The idea is that we have a list of learners with several different parameters, and if we run the program without specifying anything, it chooses the default_learner model name.

The above dict is parsed into the following pydantic BaseClass:

class Parameters(BaseModel):
    iterations: int
    size: int

class MLModel(BaseModel):
    param: Parameters

class Learners(BaseModel):
    model1: MLModel
    model2: MLModel

class Config(BaseModel):
    default_learner: str
    learners: Learners

# read in like this:
config = Config(**config_dict) # equivalent to Config.parse_obj(config_dict)

I would like to access the param field from model1 using config.default_learner with config.learners. That is, something like config.learners[config.default_learner].params. But BaseModel is not subscriptable. I've tried e.g. config.learners.dict()[config.default_learner], but that returns a dict, and I would like to keep using the nice pydantic property a.b chaining.

I could do MLModel(**config.learners.dict()[config.default_learner]).param, or even

for learner, params in config.learners:
    if learner == config.default_learner:
        param = params

but this seems rather non-ideal 🤪

Are there any better suggestions or solutions?

[thomasaarholt]

Right, I've just learnt about getattr, and realised I can modify the BaseModel to have the __get_item__ method using that, e.g.

class IndexableBaseModel(BaseModel):
    """Allows a BaseModel to return its fields by string variable indexing"""
    def __getitem__(self, item):
        return getattr(self, item)

Now if I replace all references to BaseModel in the previous example with IndexableBaseModel, config.learners[config.default_learner].params works as I expected.

Is there a rationale for why this isn't default behaviour with pydantic? Would a user expect it to function as a dict?