Object helpers

Object helpers are user-specified Python functions that can be applied to stored objects. They work similar to storage backends, that is an object helper is called upon storing (put), retrieving (get) and deleting (drop) an object to and from a store. This enables user-specified behavior for specific objects, types or kinds of objects. For example, a user may specify an object helper for a specific model type that needs customized processing before saving and upon loading (e.g. a PyTorch model).

A basic object helper

Here is a basic object helper. It instantiates a custom model class

from omegaml.backends.virtualobj import virtualobj

@virtualobj
def myhelper(obj=None, name=None, meta=None, method=None, **kwargs):
    if method == 'get':
        class MyModel:
            def predict(self, *args, **kwargs):
                return 42

        return MyModel()

om.models.put(myhelper, 'myhelper', replace=True)

Let’s use this helper with the most simple object we can store - a dictionary. We store the object as usual, specifying our myhelper object helper. When retrieving the model, our helper gets called with the method=’get’ kwarg, and returns a MyModel instance.

om.models.put({}, 'mymodel', helper='myhelper', kind='python.model', replace=True)
model = om.models.get('mymodel')
model.predict()
=> 42