skills.rest
ThomazRossito
ThomazRossitoCOMMUNITY
Explore all skills by this author →
Website𝕏@thomazrossitoGitHub

databricks-ml-training

Train and register ML models on Databricks with MLflow and Unity Catalog.

4|1|Updated May 22, 2026
Search Tags:#machine learning#model training#MLflow#databricks#feature store#Unity Catalog#XGBoost
One-click install
npx skills add https://github.com/ThomazRossito/ai-data-agents --skill databricks-ml-training-thomazrossito
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: databricks-ml-training
Source: https://github.com/ThomazRossito/ai-data-agents/tree/main/plugins/ai-data-agents/skills/databricks-ml-training
Command: npx skills add https://github.com/ThomazRossito/ai-data-agents --skill databricks-ml-training-thomazrossito

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires mlflow, optuna, xgboost, scikit-learn, databricks-feature-engineering, langgraph, databricks-langchain, databricks-agents, and includes references (resource) components.

What problem does it solve? Training ML models on Databricks involves many failure-prone details: models landing in the deprecated workspace registry instead of Unity Catalog, Optuna trials each registering a new model version, training/serving skew from inconsistent feature computation, and serverless job submission traps that silently lose outputs. This Skill encodes the canonical train-register-score flow so models are trained remotely, registered correctly, and consumed as batch UDFs or real-time endpoints. ## Core Features & Use Cases - Train and register with MLflow: XGBoost, scikit-learn, LightGBM, and PyTorch with autolog, Optuna hyperparameter search as nested runs, UC registration, and @prod/@challenger alias management. - Batch and real-time consumption: Score Delta tables via spark_udf into gold prediction tables, or hand off to model serving endpoints for sub-100ms inference. - Feature Store and Feature Views: FeatureLookup with point-in-time joins, fe.score_batch for feature-backed models, declarative Feature Views with rolling windows, Lakebase online store, and streaming Kafka features. - Custom models and agents: File-based custom PyFunc models and LangGraph ResponsesAgent with UC Function and Vector Search tools. - Use Case: Train a turbine failure classifier on silver telemetry tables as a serverless job, register it to Unity Catalog under the @prod alias, and write daily risk scores to a gold predictions table read by dashboards. ## Quick Start Use the databricks-ml-training skill to train an XGBoost classifier on my silver features table, register it to Unity Catalog, and score the latest data into a gold predictions table.

Frequently Asked Questions about databricks-ml-training

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I train and register an ML model on Databricks with MLflow?▼

Train inside a Databricks notebook submitted as a serverless job using mlflow.autolog, then call mlflow.set_registry_uri("databricks-uc") before logging so the model registers to Unity Catalog. Promote versions with set_registered_model_alias using @prod or @challenger aliases.

How to do batch scoring with a registered Databricks model?▼

Load the model with mlflow.pyfunc.spark_udf using the models:/{name}@prod URI and apply it to a Delta table, writing predictions to a gold table. For feature-store-backed models logged with fe.log_model, use fe.score_batch with key columns only instead.

Why does my model land in the workspace registry instead of Unity Catalog?▼

The MLflow registry URI defaults to the deprecated workspace registry. Call mlflow.set_registry_uri("databricks-uc") before logging any model so artifacts register to Unity Catalog with lineage and alias support.

When should I use Databricks Feature Store versus plain Delta tables?▼

Use Feature Engineering when features must be computed identically at training and serving, need point-in-time joins against labels, require sub-10ms online lookups, or are shared across models with lineage. For simple one-off models, plain UC tables with autolog are sufficient.

Can I deploy a custom LangGraph agent on Databricks?▼

Yes, implement an MLflow ResponsesAgent with LangGraph, log it with python_model pointing to the agent file, and pass resources listing the LLM endpoint, UC functions, and Vector Search indexes for passthrough auth. Deploy with agents.deploy submitted as an async serverless job.

Why does my Databricks serving endpoint return PERMISSION_DENIED?▼

The deployed model lacks credentials for resources it calls, such as UC functions, Vector Search indexes, or other endpoints. Pass a resources list with DatabricksServingEndpoint, DatabricksFunction, and DatabricksVectorSearchIndex entries to log_model for automatic passthrough authentication.

Related Skills

Explore More Data & Analytics Skills→
LaurentPRAT-DBLaurentPRAT-DB

databricks-ml

Automate end-to-end machine learning workflows on Databricks with MLflow and Unity Catalog.

Community
Advanced
andregit2026andregit2026

databricks-model-serving

Deploy and query Databricks Model Serving endpoints for ML models and GenAI agents.

Community
Intermediate
FMurrayFMurray

databricks-unity-catalog

Query Unity Catalog system tables and manage volume file operations on Databricks.

Community
Intermediate
JingyiWngJingyiWng

databricks-agent-bricks

Create and orchestrate Databricks AI agent bricks for document Q&A, SQL exploration, and multi-agent supervision.

Community
Advanced