Understand the certification
认证概述
Databricks Certified Machine Learning Professional 是 Databricks 颁发的高级机器学习认证，验证考生在 Databricks Lakehouse 平台上管理端到端 ML 生命周期的专业能力，包括特征工程、模型训练、部署和生产监控。考试共 60 道题，时间 120 分钟，通过分数线 70%。
考试领域
- MLflow 实验管理与模型注册（约 25%）：Experiment Tracking（记录 params、metrics、artifacts）；Model Registry 版本管理与阶段转换（Staging/Production/Archived）；MLflow Projects 封装；MLflow Models 的多种部署格式（pyfunc、sklearn、pytorch）
- 特征工程与 Feature Store（约 20%）：Databricks Feature Store 特征表创建与更新；Point-in-Time Lookups 防止训练/推理数据泄露；在线（Online Store）与离线特征服务；特征表与 MLflow 模型的集成注册
- 模型训练与超参数调优（约 20%）：Hyperopt 分布式超参数搜索（fmin、tpe、Trials）；AutoML 的适用场景；使用 SparkTrials 进行并行调优；交叉验证与模型评估指标
- 模型部署与服务（约 20%）：Databricks Model Serving 实时推理端点；批量推理（Spark UDF 包装 MLflow 模型）；A/B 测试与金丝雀发布；推理延迟优化；边缘部署场景
- 模型监控与漂移检测（约 15%）：数据漂移（Data Drift）与概念漂移（Concept Drift）检测方法；使用 Lakehouse Monitor 监控模型性能；触发模型再训练的策略；分布式训练（Horovod、TorchDistributor）
适合人群
- 已获得 Databricks Certified Machine Learning Associate 认证的 ML 工程师
- 负责在 Databricks 平台上构建和部署生产级机器学习系统的数据科学家
- 希望验证 MLflow 全流程与 Feature Store 专业能力的 AI/ML 从业者
Exam at a glance
- Exam code
- Databricks MLP
- Provider
- 其他
- Format
- single-choice / multiple-choice
- Language
- English
- Validity
- 3 years
Skills in focus
Learning outcomes
- 系统掌握 MLflow 端到端 ML 生命周期管理，能设计可复现、可扩展的机器学习实验和模型注册流程
- 深度理解 Databricks Feature Store 的防数据泄露设计，能构建训练与推理一致的特征工程管道
- 获得 ML Professional 认证，证明高级 MLOps 能力，支撑 Senior ML Engineer 或 MLOps 工程师职业晋升
Who this is for
- 已获得 Databricks Certified Machine Learning Associate 认证的 ML 工程师
- 负责在 Databricks 平台上构建和部署生产级机器学习系统的数据科学家
- 希望验证 MLflow 全流程与 Feature Store 专业能力的 AI/ML 从业者
- MLOps 工程师，需要掌握模型监控、漂移检测和自动化再训练策略
Before you start
- 已获得 Databricks Certified Machine Learning Associate 认证（强烈建议）
- 熟悉 MLflow 基础操作（Experiment Tracking、Model Registry）
- 具备 Python 机器学习开发经验（scikit-learn、PyTorch 或 TensorFlow）
- 了解基本的 Apache Spark 概念和 Databricks 平台使用
Your learning path
3 chapters
Plan your preparation
- Learning hours
- 12h
- With relevant experience
- 8-10 weeks
- Starting from scratch
- 12-16 weeks
- Daily study
- 1-2h
Included in your preparation
- 60+ 道专业级练习题，覆盖 Databricks 官方 ML Professional 考试所有核心领域
- 每题含深度解析，阐明 MLflow 最佳实践、Feature Store 防泄露设计、Model Serving 配置要点
- 模拟真实考试环境：120 分钟 / 60 题 / 70% 通过线
- AI 智能辅导，针对 Feature Store Point-in-Time Lookup、模型漂移检测等难点提供专项突破
- 专项覆盖 A/B 测试、金丝雀发布、分布式训练策略等 Professional 级高频考点
Preparation tips
- 60 题 90 分钟，平均每题 2 分钟，合理分配时间
- 及格分 70/100，不确定的题先标记跳过，回头再做
- 排除法非常有用 — 先排掉明显错误的选项，剩下的再分析
Common mistakes
- 没有读完所有选项就选答案 — 题目经常有"最佳"答案和"正确但不最佳"的干扰项
- 备考只刷题不理解原理 — 考试场景题需要理解底层概念
- 忽略时间管理 — 在难题上卡太久，导致后面简单题没时间做
Before you register
What does the Databricks ML Professional exam focus on?
The exam has 60 questions in 120 minutes, focusing on end-to-end ML lifecycle management: MLflow experiment tracking and model registry, feature engineering (Feature Store), model training and hyperparameter tuning, model deployment and serving, model monitoring and drift detection, and productionizing ML pipelines.
How heavily is MLflow tested in the exam?
MLflow is a core topic, accounting for roughly 25-30% of questions. You need proficiency in Experiment Tracking (logging parameters, metrics, and artifacts), Model Registry (version management and stage transitions), MLflow Projects and Models deployment formats, and using MLflow for team collaboration.
What feature engineering and Feature Store topics are covered?
You need to understand Databricks Feature Store core concepts including feature table creation and updates, Point-in-Time Lookups to prevent data leakage, online and offline feature serving, ensuring feature consistency between training and inference, and integrating feature tables with MLflow models.
What model deployment and serving topics are tested?
The exam covers multiple deployment patterns: Databricks Model Serving (real-time inference endpoints), batch inference (using Spark UDFs), and edge deployment scenarios. You need to understand model serialization formats, A/B testing, canary releases, autoscaling configuration, and optimizing inference latency and throughput.
Does the exam require deep learning knowledge?
Some deep learning fundamentals are needed. The exam covers distributed training with PyTorch/TensorFlow on Spark (Horovod/TorchDistributor), transfer learning, hyperparameter tuning (Hyperopt), and model inference with Pandas UDFs. However, it is not as deep as a dedicated deep learning certification.
How should I prepare for Databricks ML Professional?
Start by mastering ML fundamentals, then focus on end-to-end MLflow workflows and Feature Store on Databricks. Complete at least one end-to-end ML project on Databricks Community Edition. Exam questions are practice-oriented, requiring understanding of real-world best practices rather than pure theory.
Learner experiences
Loading reviews…
Done reading? Next steps
- SkillPractise Databricks ML Professional by exam domainQuestions are grouped by the official exam domains, so you can see which domain still needs work.Open the question bank →
- SkillDatabricks ML Professional learning pathWork through the exam scope chapter by chapter, then come back to the questions.See the learning path →
Start with the first chapter.
2 free chapters · Full preparation from $39