CODEWITHSUNDEEP
mlflow
Nastasia
Nastasia

Reputation: 657

How to load a model using the object "mlflow.tracking.client.MlflowClient"?

I'm stuck with the MLFlow model registry. Does anyone know how to load a model using the object "mlflow.tracking.client.MlflowClient"?

I would like to do a predict after with that. I'm sure I'm wrong somewhere because I've already done that in the past. I'm not able to find it in the doc, in the web.

Upvotes: 0

Views: 2259

Answers (1)

Bram
Bram

Reputation: 396

You'll have to make use of mlflow.<model_flavor>.load_model() to load a given model from the Model Registry. For example:

import mlflow.pyfunc

model = mlflow.pyfunc.load_model(
          model_uri="models:/<model_name>/<model_version>"
          )

model.predict(...)

With mlflow.tracking.client.MlflowClient you can retrieve metadata about a model from the model registry, but for retrieving the actual model you will need to use mlflow.<model_flavor>.load_model. For example, you could use the MlflowClient to get the download URI for a given model, and then use mlflow.<flavor>.load_model to retrieve that model.

model_uri = client.get_model_version_download_uri("<model_name>", <version>)
model = mlflow.pyfunc.load_model(model_uri)

model.predict(...)

Upvotes: 2

Related Questions