Edit on GitHub

Working with models

To be able to use all MLEM features, you need to turn your model into a MLEM model first.

The easiest way to do this is to use save API method.

from mlem.api import save

save(model, "models/mymodel", sample_data=df)

After that you could work with your model from CLI or call API methods, passing down a path to the saved model.

model object can be any supported python object. MLEM works with all major popular machine learning frameworks as well as arbitrary python callables.

You can find full list here as well as additional documentation.

For most usecases it's mandatory to provide sample_data argument. Typically, it is the same data object you provide for your model .predict method.

MLEM uses it to infer your model's signature which will be needed to build and deploy it later.

Additionally, MLEM will automatically infer your model requirements (including any local code). The determined requirements can be used to create virtual environments

Alternative ways to create MLEM model

Existing model files can be imported.

You can also create MLEM model from your object without saving it. Use .dump to save it later.

from mlem.core.objects import MlemModel

mlem_model = MlemModel.from_obj(model, sample_data=df)
mlem_model.dump("models/mymodel")

This may be useful if you're going to work with MLEM model from API inside the same Python process, or you want to persist it somewhere but would like to avoid re-reading it after mlem.api.save.

Besides, note that you can pass both model and mlem_model to MLEM Python API methods. Passing mlem_model make sense when you're going to use MLEM with it model multiple times, and want to save some time on creating a MLEM model object from scratch each time.

Content

­čÉŤ Found an issue? Let us know! Or fix it:

Edit on GitHub

ÔŁô Have a question? Join our chat, we will help you:

Discord Chat