AllOnIAMLOps Library

This module is designed to make model serving on the Aleia platform as seamless as possible.

Seldon

We provide an encapsulation of the Seldon MLOps framework to easily deploy and interact with AI models, either off-the-shelf or trained using the aleia_model module.

Example: after having trained a model called titanic, the user wants to deploy it and be able to make predictions through a REST endpoint.

import alloniamlops

ret = alloniamlops.seldon.deploy_model(
    "model_titanic", # model name
    1,               # revision number
)

This snippet will deploy the model using Seldon and return its URL.

The user can also list the currently deployed models:

import alloniamlops

aleiamlops.seldon.list_model()

Using the URL of a deployed model, it is easy to perform a prediction:

from allonias3 import S3Path
import alloniamlops

df = S3Path("dataset/dataset.csv").read()
df.drop("Gender", axis=1, inplace=True)

alloniamlops.seldon.model_predict(
    seldon_deployment_url,
    names=df.columns.values,
    data=df.to_numpy(),
    payload_type="ndarray",
    debug=True,
)

If the results are unexpected, logs can be examined as follow:

import alloniamlops

aleiamlops.seldon.tail(seldon_deployment_id, last_lines=20)

Finally, the user may want to shutdown the instance running the model:

import alloniamlops

aleiamlops.seldon.delete_model_deployment(seldon_deployment_id)