Execute Databricks secondary workflow: MLflow model training and deployment.
Use when building ML pipelines, training models, or deploying to production.
Trigger with phrases like "databricks ML", "mlflow training",
"databricks model", "feature store", "model registry".
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Execute Databricks secondary workflow: MLflow model training and deployment.
Use when building ML pipelines, training models, or deploying to production.
Trigger with phrases like "databricks ML", "mlflow training",
"databricks model", "feature store", "model registry".
Databricks Core Workflow B: MLflow Training & Serving
Overview
Full ML lifecycle on Databricks: Feature Engineering Client for discoverable features, MLflow experiment tracking with auto-logging, Unity Catalog model registry with aliases (champion/challenger), and Mosaic AI Model Serving endpoints for real-time inference via REST API.
Prerequisites
Completed databricks-install-auth and databricks-core-workflow-a
databricks-sdk, mlflow, scikit-learn installed
Unity Catalog enabled (required for model registry)
Instructions
Step 1: Feature Engineering with Feature Store
Create a feature table in Unity Catalog so features are discoverable and reusable.
from databricks.feature_engineering import FeatureEngineeringClient
from pyspark.sql import SparkSession
import pyspark.sql.functions as F
spark = SparkSession.builder.getOrCreate()
fe = FeatureEngineeringClient()
# Build features from gold layer tables
user_features = (
spark.table("prod_catalog.gold.user_events")
.groupBy("user_id")
.agg(
F.count("event_id").alias("total_events"),
F.avg("session_duration_sec").alias("avg_session_sec"),
F.max("event_timestamp").alias("last_active"),
F.countDistinct("event_type").alias("unique_event_types"),
F.datediff(F.current_date(), F.max("event_timestamp")).alias("days_since_last_active"),
)
)
# Register as a feature table (creates or updates)
fe.create_table(
name="prod_catalog.ml_features.user_behavior",
primary_keys=["user_id"],
df=user_features,
description="User behavioral features for churn prediction",
)
Step 2: MLflow Experiment Tracking
import mlflow
from sklearn.model_selection import train_test_split
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
# Point MLflow to Databricks tracking server
mlflow.set_tracking_uri("databricks")
mlflow.set_experiment("/Users/team@company.com/churn-prediction")
# Load features
features_df = spark.table("prod_catalog.ml_features.user_behavior").toPandas()
X = features_df.drop(columns=["user_id", "churned"])
y = features_df["churned"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
# Train with experiment trackingwith mlflow.start_run(run_name="gbm-baseline") as run:
params = {"n_estimators": 200, "max_depth": 5, "learning_rate": 0.1}
mlflow.log_params(params)
model = GradientBoostingClassifier(**params)
model.fit(X_train, y_train)
y_pred = model.predict(X_test)
metrics = {
"accuracy": accuracy_score(y_test, y_pred),
"precision": precision_score(y_test, y_pred),
"recall": recall_score(y_test, y_pred),
"f1": f1_score(y_test, y_pred),
}
mlflow.log_metrics(metrics)
# Log model with signature for serving validation
mlflow.sklearn.log_model(
model,
artifact_path="model",
input_example=X_test.iloc[:5],
registered_model_name="prod_catalog.ml_models.churn_predictor",
)
print(f"Run {run.info.run_id}: accuracy={metrics['accuracy']:.3f}")
Step 3: Model Registry with Aliases
Unity Catalog model registry replaces legacy stages with aliases (champion, challenger).
from mlflow import MlflowClient
client = MlflowClient()
model_name = "prod_catalog.ml_models.churn_predictor"# List versionsfor mv in client.search_model_versions(f"name='{model_name}'"):
print(f"v{mv.version}: status={mv.status}, aliases={mv.aliases}")
# Promote best version to champion
client.set_registered_model_alias(model_name, alias="champion", version="3")
# Load model by alias in downstream code
champion = mlflow.pyfunc.load_model(f"models:/{model_name}@champion")
predictions = champion.predict(X_test)
Step 4: Deploy Model Serving Endpoint
Mosaic AI Model Serving creates a REST API endpoint with auto-scaling.
from databricks.sdk import WorkspaceClient
from databricks.sdk.service.serving import (
EndpointCoreConfigInput, ServedEntityInput,
)
w = WorkspaceClient()
# Create or update a serving endpoint
endpoint = w.serving_endpoints.create_and_wait(
name="churn-predictor-prod",
config=EndpointCoreConfigInput(
served_entities=[
ServedEntityInput(
entity_name="prod_catalog.ml_models.churn_predictor",
entity_version="3",
workload_size="Small",
scale_to_zero_enabled=True,
)
]
),
)
print(f"Endpoint ready: {endpoint.name} ({endpoint.state.ready})")
