Google Machine Learning and Generative AI for Solutions Architects

Una inmersión más profunda

English version here!

Bueno, he dedicado un tiempo a “Google Machine Learning and Generative AI for Solutions Architects” y debo decir que es un trabajo bastante impresionante. Todos sabemos que hay toneladas de libros sobre inteligencia artificial, machine learning y la computación en la nube, pero este es diferente. Lo digo porque no trata de repetir la misma vieja teoría; esta obra es como un manual de campo para construir sistemas de IA reales y listos para llevar a producción en Google Cloud.

Es como un plano del mundo real, no solo un libro de texto

Lo que realmente distingue a este libro es su enfoque práctico. No solo habla de inteligencia artificial y machine learning; te muestra cómo hacerlo. Te guía a través de todo el ciclo de vida, desde la gestión de datos desordenados hasta la implementación y el monitoreo de modelos a escala. No se trata solo de algoritmos y matemáticas; se trata de los aspectos prácticos de la creación de soluciones de inteligencia artificial que realmente funcionan en el mundo real. ¿Y los ejercicios prácticos? No son solo teorías superficiales; son proyectos reales que puedes desarrollar y de los que puedes aprender.

Google Machine Learning and Generative AI for Solutions Architects + Este humilde servidor

Aquí es donde realmente brilla

• MLOps está al frente y en el centro: No se trata de una idea de último momento. El libro dedica una parte importante a MLOps y abarca todo, desde la automatización hasta la supervisión y los procesos de CI/CD. Está claro que el autor entiende que crear un modelo es solo la mitad de la batalla; el verdadero desafío es mantenerlo funcionando sin problemas en producción.
• La IA generativa recibe el reconocimiento que se merece: No se trata solo del machine learning tradicional. El libro profundiza en el mundo de la IA generativa y abarca los embeddings, las bases de datos vectoriales, RAG e incluso LangChain. Es una excelente manera de ponerse al día con las últimas tendencias.
• Desafíos del mundo real abordados: No elude las partes complicadas de la IA y el machine learning. Aborda los problemas comunes que todos enfrentamos, como la calidad de los datos, el sesgo, la desviación del modelo y la seguridad. No se trata solo de la tecnología, sino de los problemas del mundo real que debemos resolver.
• Experiencia en Google Cloud: Es una inmersión profunda en el ecosistema de IA y ML de Google Cloud. Obtendrás una comprensión sólida de sus servicios, cómo funcionan juntos y cuándo usar cada uno. Es como tener un experto de Google Cloud que te guíe a través del proceso.
• Es un 3 en 1: Es como obtener tres cursos en uno: una base sólida en IA/ML, una guía completa de Google Cloud y un manual práctico para crear soluciones de nivel empresarial.

Table of Contents

Preface

Part 1: The Basics

1. AI/ML Concepts, Real-World Applications, and Challenges

• Terminology — AI, ML, DL, and GenAI
• A brief history of AI/ML
• AI/ML and cloud computing
• ML approaches and use cases
• Supervised learning (SL)
• Unsupervised learning (UL)
• Reinforcement learning (RL)
• A brief discussion of ML basic concepts
• Linear algebra
• Calculus
• Statistics and probability
• Metrics
• Common challenges in developing ML applications
• Gathering, processing, and labeling data
• Organizational challenges
• Operationalization and ongoing management of AI/ML models
• Edge cases
• Summary

2. Understanding the ML Model Development Life Cycle

• An overview of the ML model development life cycle
• SDLC — a quick recap
• Typical ML project stages
• Roles and personas in AI/ML projects
• Common challenges encountered in the ML model development life cycle
• Finding and gathering relevant data
• Picking an algorithm and model architecture
• Data labeling
• Training models
• Configuring and tuning hyperparameters
• Evaluating models
• Deploying models
• Monitoring models after deployment
• Best practices for overcoming common challenges
• Finding and gathering relevant data
• Data labeling
• Picking an algorithm and model architecture
• Training models
• Configuring and tuning hyperparameters
• Deploying models
• Monitoring models after deployment
• Summary

3. AI/ML Tooling and the Google Cloud AI/ML Landscape

• Why Google Cloud?
• Prerequisites for using Google Cloud tools and services
• Security, privacy, and compliance
• Google Cloud services overview
• Google Cloud computing services
• Google Cloud integration services
• Networking and connectivity
• Google Cloud tools for data storage and processing
• Data ingestion
• Data storage
• Data management
• Data processing
• Google Cloud AI tools and AutoML
• NLP
• The Natural Language API
• Computer vision
• AutoML
• Google Cloud Vertex AI
• Standard industry tools on Google Cloud
• Choosing the right tool for the job
• Summary

Part 2: Diving in and building AI/ML solutions

4. Utilizing Google Cloud’s High-Level AI Services

• Prerequisites for this chapter
• Cloning this book’s GitHub repository to your local machine
• The Google Cloud console
• Google Cloud project
• Google Cloud Billing
• Google Cloud Shell
• Authentication
• Enabling the relevant Google Cloud APIs
• Storing authentication credentials in environment variables
• Creating a directory and cloning our GitHub repository
• Detecting text in images with the Cloud Vision API
• Using Document AI to extract information from documents
• Document AI concepts
• Performing OCR with Document AI
• Using the Google Cloud Natural Language API to get sentiment analysis insights from textual inputs
• Sentiment analysis with the Natural Language API
• Classifying content with the Natural Language API
• Using Vertex AI AutoML
• Use case — forecasting
• Summary

5. Building Custom ML Models on Google Cloud

• Background information — libraries
• scikit-learn
• Matplotlib
• pandas
• XGBoost
• Prerequisites for this chapter
• Vertex AI Workbench
• Creating a Vertex AI Workbench instance
• Vertex AI Notebook JupyterLab integrations
• Cloning the GitHub repository
• UML with scikit-learn on Vertex AI
• K-means
• Implementing a UML workload in Vertex AI
• Implementing a regression model with scikit-learn on Vertex AI
• Implementing a classification model with XGBoost on Vertex AI
• Decision trees
• Summary

6. Diving Deeper — Preparing and Processing Data for AI/ML Workloads on Google Cloud

• Prerequisites for this chapter
• Enabling APIs
• IAM permissions
• Cloud Storage bucket folders
• Uploading data
• Fundamental concepts in this chapter
• Ingesting data into Google Cloud
• ETL and ELT
• Batch and streaming data processing
• Data pipelines
• Exploring, visualizing, and preparing data
• Batch data pipelines
• Batch data pipeline concepts and tools
• Building our batch data pipeline
• Cloud Composer
• Google Cloud Serverless Spark
• Streaming data pipelines
• Streaming data pipeline concepts and tools
• Building our streaming data pipeline
• Creating a BigQuery dataset
• Creating a BigQuery table
• Creating a Dataflow job from a template
• Verifying data in BigQuery
• Creating a Dataflow notebook
• Verifying data in BigQuery
• Summary

7. Feature Engineering and Dimensionality Reduction

• Fundamental concepts in this chapter
• Dimensions and features
• Overfitting, underfitting, and regularization
• Feature selection and feature engineering
• The curse of dimensionality
• Dimensionality reduction
• t-SNE
• Using PCA and LDA for dimensionality reduction
• Feature engineering
• Vertex AI Feature Store
• Introduction to Vertex AI Feature Store
• Online versus offline feature serving
• Building our feature store
• How features are used during online inference
• Summary

8. Hyperparameters and Optimization

• Prerequisites
• Enabling the Artifact Registry API
• Creating an AI/ML service account
• Concepts
• Model evaluation metrics used in this chapter
• What are hyperparameters?
• Hyperparameter optimization
• Methods for optimizing hyperparameter values
• Hands-on: performing hyperparameter tuning in Vertex AI
• Vertex AI Vizier
• Summary

9. Neural Networks and Deep Learning

• NN and DL concepts
• Neurons and the perceptron
• Backpropagation
• Activation functions
• Libraries
• TensorFlow
• Keras
• Implementing an MLP in TensorFlow
• NN architectures, challenges, and optimization
• Common NN architectures
• Common NN challenges
• Summary

10. Deploying, Monitoring, and Scaling in Production

• How do I make my models available to my applications?
• Fundamental concepts for serving models
• Online and offline model serving
• Vertex AI Model Registry
• Vertex AI prediction service
• A/B testing
• Common challenges of serving models in production
• Deployment infrastructure
• Model availability and scaling in production
• Data quality
• Model/data/concept drift
• Security and privacy
• Model interpretability
• Tracking ML model metadata
• Integration with existing systems
• Monitoring
• Monitoring models in production
• Objective model performance
• Monitoring specifically for data drift
• Anomalous model behavior
• Resource utilization
• Model bias and fairness
• Addressing model performance degradation
• Optimizing for AI/ML at the edge
• Model optimization
• Optimization beyond model techniques
• Summary

11. Machine Learning Engineering and MLOps with Google Cloud

• An introduction to MLOps
• Why MLOps is needed for deploying large-scale ML workloads
• Model management and versioning
• Productivity and automation
• Reproducibility
• Continuous Integration/Continuous Deployment (CI/CD)
• Model validation and testing
• Monitoring and maintenance
• Collaboration and communication
• Regulatory compliance and governance
• MLOps tools
• Pipeline orchestration tools
• Experiment and lineage tracking tools
• Model deployment and monitoring tools
• Model interpretability and explainability tools
• Implementing MLOps on Google Cloud using Vertex AI Pipelines
• Prerequisite: IAM permissions
• Implementation
• Summary

12. Bias, Explainability, Fairness, and Lineage

• An overview of bias, explainability, and fairness in AI/ML
• Bias
• Fairness
• Explainability
• How to detect and mitigate bias in datasets
• Data exploration and visualization
• Specific tools for detecting dependencies between features
• Mechanisms incorporating model prediction results
• Using explainability to understand ML models and reduce bias
• Explainability techniques, methods, and tools
• Feature importance
• Reducing bias and enhancing fairness
• Additional libraries
• The importance of lineage tracking in ML model development
• ML metadata service terminology
• Lineage tracking in Vertex AI
• Summary

13. ML Governance and the Google Cloud Architecture Framework

• ML governance
• Data governance
• ML model governance
• Operationalization of ML governance
• ML governance in different industries and locations
• Keeping up with the evolving landscape of ML governance
• An overview of the Google Cloud Architecture Framework
• Pillar 1 — System design
• Pillar 2 — Operational excellence
• Pillar 3 — Security, privacy, and compliance
• Pillar 4 — Reliability
• Pillar 5 — Cost optimization
• Pillar 6 — Performance optimization
• Architecture Framework concepts about AI/ML workloads on Google Cloud
• Data collection and preparation
• Model building and training
• Model evaluation and deployment
• Summary

14. Additional AI/ML Tools, Frameworks, and Considerations

• Prerequisite topics and steps
• Staging files for serverless Spark MLlib activities
• BQML
• Using BQML
• When to use BQML versus other tools for AI/ML use cases
• BigQuery Studio
• Hardware considerations for AI/ML workloads
• CPUs, GPUs, and TPUs
• Additional open source tools and frameworks –Spark MLlib, Ray, and PyTorch on Google Cloud
• Spark MLlib
• Ray
• PyTorch
• Large-scale distributed model training
• Data parallelism and model parallelism
• Distributed training update process
• Federated learning
• Transitioning to Generative AI
• CNNs and computer vision
• RNNs, LSTMs, and transformers
• Summary

Part 3: Generative AI

15. Introduction to Generative AI

• Fundamentals of GenAI
• What is GenAI?
• What is non-GenAI?
• Diving deeper into GenAI versus non-GenAI
• GenAI techniques and evolution
• Embeddings and latent space
• LLMs
• Evolution of LLMs
• Building LLMs
• Summary

16. Advanced Generative AI Concepts and Use Cases

• Advanced tuning and optimization techniques
• Prompt engineering
• Transfer Learning (TL)
• Aligning with human values and expectations
• Embeddings and vector databases
• Embeddings and similarity of concepts
• Vector databases
• How vector databases work
• Creating embeddings
• Retrieval-Augmented Generation (RAG)
• How RAG works
• Multimodal models
• Why multimodality matters
• Multimodality challenges
• GenAI model evaluation
• Human evaluation
• BLEU
• ROUGE
• IS
• FID
• Auto-raters and side-by-side evaluations
• LangChain
• Summary

17. Generative AI on Google Cloud

• Overview of generative AI in Google Cloud
• Google’s generative AI models
• Open source and third-party generative AI models on Google Cloud
• Vector databases in Google Cloud
• A detailed exploration of Google Cloud generative AI
• Google Cloud Vertex AI Studio
• A detailed exploration of Google Cloud vector database options
• Implementing generative AI solutions on Google Cloud
• Building a Vertex AI Search and Conversation application
• Summary

18. Bringing It All Together: Building ML Solutions with Google Cloud and Vertex AI

• Prerequisites
• Using the Google Cloud Shell
• Building a RAG implementation piece by piece
• Tokens and chunks
• Example RAG solution architecture
• Building the RAG implementation on Google Cloud
• An example business use case and reference architecture
• Additional background on our use case
• The reference architecture
• Building and implementing the use case on Google Cloud
• Recap and next steps
• Next steps

Index

Other Books You May Enjoy

¿Quién debería adquirir una copia?

Si eres un arquitecto de soluciones, o estás en ese camino, este libro es imprescindible. También es un gran recurso para los científicos de datos que quieren ir más allá del laboratorio y entrar en producción. No es un libro para principiantes, pero hace un buen trabajo al explicar conceptos complejos con claridad, por lo que no necesitas ser un genio de las matemáticas para seguirlo.

Veredicto Final

Este libro es un recurso fantástico para cualquier persona que se tome en serio la creación e implementación de soluciones de inteligencia artificial y de machine learning en Google Cloud. Es práctico, completo y te ofrece una perspectiva del mundo real sobre lo que se necesita para tener éxito. No se trata solo de la tecnología; se trata de todo el proceso, de principio a fin. Si buscas llevar tus habilidades de AI/ML al siguiente nivel y crear soluciones que realmente brinden valor comercial, te recomiendo que lo consultes. No es solo un libro; es una inversión en tu futuro.

Gracias por leer. ¿Qué tal un videito?

¡Dale 50 veces el botón 👏 y comenta si quieres que haga más reseñas de libros!

¡Sigue a David Regalado para obtener más contenido educativo!