Add Feast feature store example

[tmvfb]

No description provided.

[hsteude]

The pip install here needs the redis extra, otherwise materialization to Redis fails at runtime. Should be:

!pip install -q "feast[redis]" scikit-learn mlflow

Tested on perfect-shastasaurus — without [redis], materialize crashes with a missing dependency error.

[hsteude]

This produces a DeprecationWarning:

DeprecationWarning: Entity value_type will be mandatory in the next release. Please specify a value_type for entity 'driver_id'.

Easy fix — add value_type:

from feast.value_type import ValueType

driver = Entity(
    name="driver_id",
    description="Unique driver identifier",
    value_type=ValueType.INT64,
)

[hsteude]

This fails with RestException: INTERNAL_ERROR: Response: {'detail': 'Authentication required'} because MLflow on prokube requires authentication via the OIDC proxy.

On prokube notebooks, MLflow access is normally pre-configured via a PodDefault that injects the correct MLFLOW_TRACKING_URI and auth token. Setting the tracking URI manually to the internal service bypasses the auth layer.

This whole block should probably be replaced with just:

import mlflow

mlflow.set_experiment("feast-driver-prediction")

...and rely on the environment variable MLFLOW_TRACKING_URI that the PodDefault already sets. If users don't have the MLflow PodDefault enabled, the notebook should mention that as a prerequisite.

[hsteude]

This fails with RestException: INTERNAL_ERROR: Response: {'detail': 'Authentication required'} because MLflow on prokube requires authentication via the OIDC proxy.

On prokube notebooks, MLflow access is normally pre-configured via a PodDefault that injects the correct MLFLOW_TRACKING_URI and auth token. Setting the tracking URI manually to the internal service bypasses the auth layer.

This whole block should probably be replaced with just:

import mlflow

mlflow.set_experiment("feast-driver-prediction")

...and rely on the environment variable MLFLOW_TRACKING_URI that the PodDefault already sets. If users don't have the MLflow PodDefault enabled, the notebook should mention that as a prerequisite.

pure ai slop, but it did not work and i really got an auth error :(

[tmvfb]

lol where the hell did it decide to get MLflow from

AI can't test a proper MLflow setup without my intermission of course (I need to give it a token)

Haven't tested (and even looked at this!) myself, will take the effort on Monday and clean this up, sorry for my AI slop haha