Official Google Cloud Certified Professional Machine Learning Engineer Study Guide

ABOUT THE AUTHORS MONA is an AI/ML specialist in the Google Public Sector. She is the author of Natural Language Processing with AWS AI Services and a frequent speaker at cloud computing and machine learning events. She was a Sr. AI/ML specialist SA at AWS before joining Google. She has 14 Certifications and has created courses for AWS AI/ML Certification Speciality Exam readiness. She has authored 17 articles on AI/ML and also co-authored a research paper on CORD-19 Neural Search, which won an award at the AAAI Conference on Artificial Intelligence Pratap Ramamurthy is an AI/ML Specialist Customer Engineer in Google Cloud. Previously, he worked as a Sr. Principal Solution Architect at H2O.ai and before that was a Partner Solution Architect at AWS. He has authored several research papers and holds 3 patents.

• Introduction xxiAssessment TestxxxiiChapter 1 Framing ML Problems 1Translating Business Use Cases 3Machine Learning Approaches 5Supervised, Unsupervised, and Semi- supervised Learning 5Classification, Regression, Forecasting, and Clustering 7ML Success Metrics 8Regression 12Responsible AI Practices 13Summary 14Exam Essentials 14Review Questions 15Chapter 2 Exploring Data and Building Data Pipelines 19Visualization 20Box Plot 20Line Plot 21Bar Plot 21Scatterplot 22Statistics Fundamentals 22Mean 22Median 22Mode 23Outlier Detection 23Standard Deviation 23Correlation 24Data Quality and Reliability 24Data Skew 25Data Cleaning 25Scaling 25Log Scaling 26Z-score 26Clipping 26Handling Outliers 26Establishing Data Constraints 27Exploration and Validation at Big- Data Scale 27Running TFDV on Google Cloud Platform 28Organizing and Optimizing Training Datasets 29Imbalanced Data 29Data Splitting 31Data Splitting Strategy for Online Systems 31Handling Missing Data 32Data Leakage 33Summary 34Exam Essentials 34Review Questions 36Chapter 3 Feature Engineering 39Consistent Data Preprocessing 40Encoding Structured Data Types 41Mapping Numeric Values 42Mapping Categorical Values 42Feature Selection 44Class Imbalance 44Classification Threshold with Precision and Recall 45Area under the Curve (AUC) 46Feature Crosses 46TensorFlow Transform 49TensorFlow Data API (tf.data) 49TensorFlow Transform 49GCP Data and ETL Tools 51Summary 51Exam Essentials 52Review Questions 53Chapter 4 Choosing the Right ML Infrastructure 57Pretrained vs. AutoML vs. Custom Models 58Pretrained Models 60Vision AI 61Video AI 62Natural Language AI 62Translation AI 63Speech- to- Text 63Text- to- Speech 64AutoML 64AutoML for Tables or Structured Data 64AutoML for Images and Video 66AutoML for Text 67Recommendations AI/Retail AI 68Document AI 69Dialogflow and Contact Center AI 69Custom Training 70How a CPU Works 71GPU 71TPU 72Provisioning for Predictions 74Scaling Behavior 75Finding the Ideal Machine Type 75Edge TPU 76Deploy to Android or iOS Device 76Summary 77Exam Essentials 77Review Questions 78Chapter 5 Architecting ML Solutions 83Designing Reliable, Scalable, and Highly Available ml Solutions 84Choosing an Appropriate ML Service 86Data Collection and Data Management 87Google Cloud Storage (GCS) 88BigQuery 88Vertex AI Managed Datasets 89Vertex AI Feature Store 89NoSQL Data Store 90Automation and Orchestration 91Use Vertex AI Pipelines to Orchestrate the ML Workflow 92Use Kubeflow Pipelines for Flexible Pipeline Construction 92Use TensorFlow Extended SDK to Leverage Pre-built Components for Common Steps 93When to Use Which Pipeline 93Serving 94Offline or Batch Prediction 94Online Prediction 95Summary 97Exam Essentials 97Review Questions 98Chapter 6 Building Secure ML Pipelines 103Building Secure ML Systems 104Encryption at Rest 104Encryption in Transit 105Encryption in Use 105Identity and Access Management 105IAM Permissions for Vertex AI Workbench 106Securing a Network with Vertex AI 109Privacy Implications of Data Usage and Collection 113Google Cloud Data Loss Prevention 114Google Cloud Healthcare API for PHI Identification 115Best Practices for Removing Sensitive Data 116Summary 117Exam Essentials 118Review Questions 119Chapter 7 Model Building 121Choice of Framework and Model Parallelism 122Data Parallelism 122Model Parallelism 123Modeling Techniques 125Artificial Neural Network 126Deep Neural Network (DNN) 126Convolutional Neural Network 126Recurrent Neural Network 127What Loss Function to Use 127Gradient Descent 128Learning Rate 129Batch 129Batch Size 129Epoch 129Hyperparameters 129Transfer Learning 130Semi-supervised Learning 131When You Need Semi-supervised Learning 131Limitations of SSL 131Data Augmentation 132Offline Augmentation 132Online Augmentation 132Model Generalization and Strategies to Handle Overfitting and Underfitting 133Bias Variance Trade- Off 133Underfitting 133Overfitting 134Regularization 134Summary 136Exam Essentials 137Review Questions 138Chapter 8 Model Training and Hyperparameter Tuning 143Ingestion of Various File Types into Training 145Collect 146Process 147Store and Analyze 150Developing Models in Vertex AI Workbench by Using Common Frameworks 151Creating a Managed Notebook 153Exploring Managed JupyterLab Features 154Data Integration 155BigQuery Integration 155Ability to Scale the Compute Up or Down 156Git Integration for Team Collaboration 156Schedule or Execute a Notebook Code 158Creating a User-Managed Notebook 159Training a Model as a Job in Different Environments 161Training Workflow with Vertex AI 162Training Dataset Options in Vertex AI 163Pre-built Containers 163Custom Containers 166Distributed Training 168Hyperparameter Tuning 169Why Hyperparameters Are Important 170Techniques to Speed Up Hyperparameter Optimization 171How Vertex AI Hyperparameter Tuning Works 171Vertex AI Vizier 174Tracking Metrics During Training 175Interactive Shell 175TensorFlow Profiler 177What-If Tool 177Retraining/Redeployment Evaluation 178Data Drift 178Concept Drift 178When Should a Model Be Retrained? 178Unit Testing for Model Training and Serving 179Testing for Updates in API Calls 180Testing for Algorithmic Correctness 180Summary 180Exam Essentials 181Review Questions 182Chapter 9 Model Explainability on Vertex AI 187Model Explainability on Vertex AI 188Explainable AI 188Interpretability and Explainability 189Feature Importance 189Vertex Explainable AI 189Data Bias and Fairness 193ML Solution Readiness 194How to Set Up Explanations in the Vertex AI 195Summary 196Exam Essentials 196Review Questions 197Chapter 10 Scaling Models in Production 199Scaling Prediction Service 200TensorFlow Serving 201Serving (Online, Batch, and Caching) 203Real- Time Static and Dynamic Reference Features 203Pre-computing and Caching Prediction 206Google Cloud Serving Options 207Online Predictions 207Batch Predictions 212Hosting Third- Party Pipelines (MLFlow) on Google Cloud 213Testing for Target Performance 214Configuring Triggers and Pipeline Schedules 215Summary 216Exam Essentials 217Review Questions 218Chapter 11 Designing ML Training Pipelines 221Orchestration Frameworks 223Kubeflow Pipelines 224Vertex AI Pipelines 225Apache Airflow 228Cloud Composer 229Comparison of Tools 229Identification of Components, Parameters, Triggers, and Compute Needs 230Schedule the Workflows with Kubeflow Pipelines 230Schedule Vertex AI Pipelines 232System Design with Kubeflow/TFX 232System Design with Kubeflow DSL 232System Design with TFX 234Hybrid or Multicloud Strategies 235Summary 236Exam Essentials 237Review Questions 238Chapter 12 Model Monitoring, Tracking, and Auditing Metadata 241Model Monitoring 242Concept Drift 242Data Drift 243Model Monitoring on Vertex AI 243Drift and Skew Calculation 244Input Schemas 245Logging Strategy 247Types of Prediction Logs 247Log Settings 248Model Monitoring and Logging 248Model and Dataset Lineage 249Vertex ML Metadata 249Vertex AI Experiments 252Vertex AI Debugging 253Summary 253Exam Essentials 254Review Questions 255Chapter 13 Maintaining ML Solutions 259MLOps Maturity 260MLOps Level 0: Manual/Tactical Phase 261MLOps Level 1: Strategic Automation Phase 263MLOps Level 2: CI/CD Automation, Transformational Phase 264Retraining and Versioning Models 266Triggers for Retraining 267Versioning Models 267Feature Store 268Solution 268Data Model 269Ingestion and Serving 269Vertex AI Permissions Model 270Custom Service Account 270Access Transparency in Vertex AI 271Common Training and Serving Errors 271Training Time Errors 271Serving Time Errors 271TensorFlow Data Validation 272Vertex AI Debugging Shell 272Summary 272Exam Essentials 273Review Questions 274Chapter 14 BigQuery ML 279BigQuery – Data Access 280BigQuery ML Algorithms 282Model Training 282Model Evaluation 284Prediction 285Explainability in BigQuery ML 286BigQuery ML vs. Vertex AI Tables 289Interoperability with Vertex AI 289Access BigQuery Public Dataset 289Import BigQuery Data into Vertex AI 290Access BigQuery Data from Vertex AI Workbench Notebooks 290Analyze Test Prediction Data in BigQuery 290Export Vertex AI Batch Prediction Results 290Export BigQuery Models into Vertex AI 291BigQuery Design Patterns 291Hashed Feature 291Transforms 291Summary 292Exam Essentials 293Review Questions 294Appendix Answers to Review Questions 299Chapter 1: Framing ML Problems 300Chapter 2: Exploring Data and Building Data Pipelines 301Chapter 3: Feature Engineering 302Chapter 4: Choosing the Right ML Infrastructure 302Chapter 5: Architecting ML Solutions 304Chapter 6: Building Secure ML Pipelines 305Chapter 7: Model Building 306Chapter 8: Model Training and Hyperparameter Tuning 307Chapter 9: Model Explainability on Vertex AI 308Chapter 10: Scaling Models in Production 308Chapter 11: Designing ML Training Pipelines 309Chapter 12: Model Monitoring, Tracking, and Auditing Metadata 310Chapter 13: Maintaining ML Solutions 311Chapter 14: BigQuery ML 313Index 315