Deploying a XGBoost model

As mentioned in the deploying models guide, deploying models via Verta Inference is a two step process: (1) first create an endpoint, and (2) update the endpoint with a model.

This tutorial explains how Verta Inference can be used to deploy an XGBoost model.

1. Create an endpoint

Users can create an endpoint using Client.create_endpoint() as follows:

wine_endpoint = client.create_endpoint(path="/wine")

2. Updating the endpoint with a RMV

As discussed in the Catalog Overview, there are multiple ways to create an RMV for a XGBoost model.

Verta supports directly logging and deploying a model object from XGBoost's Scikit-Learn API. The XGBoost convenience function can be used to create a Verta Standard Model:

Note that in addition to xgboost, this also requires scikit-learn as a dependency for deployment.

from verta.environment import Python

model_version = registered_model.create_standard_model_from_xgboost(
    model,
    environment=Python(requirements=["xgboost", "scikit-learn"]),
    name="v1",
)

Alternatively, an XGBoost saved model can be used as an artifact in a class that extends VertaModelBase, which is useful for deploying XGBoost's typical Booster models:

from verta.environment import Python
from verta.registry import VertaModelBase, verify_io
from xgboost import Booster, DMatrix


class Model(VertaModelBase):
    def __init__(self, artifacts):
        self.model = Booster()
        self.model.load_model(artifacts["xgb_model"])

    @verify_io
    def predict(self, input):
        input = DMatrix(input)
        output = self.model.predict(input)
        return output.tolist()


model_version = registered_model.create_standard_model(
    Model,
    artifacts={"xgb_model": "path/to/model.json"},
    environment=Python(requirements=["xgboost"]),
    name="v1",
)

Regardless of how a Registered Model Version has been created, the endpoint defined above can now be upated and we can make predictions against it.

wine_endpoint = client.get_or_create_endpoint("wine")
wine_endpoint.update(model_version, wait=True)
deployed_model = wine_endpoint.get_deployed_model()
deployed_model.predict([x_test[0]])

The full code for this tutorial can be found here.