Book Image

Mastering Azure Machine Learning - Second Edition

By : Christoph Körner, Marcel Alsdorf
Book Image

Mastering Azure Machine Learning - Second Edition

By: Christoph Körner, Marcel Alsdorf

Overview of this book

Azure Machine Learning is a cloud service for accelerating and managing the machine learning (ML) project life cycle that ML professionals, data scientists, and engineers can use in their day-to-day workflows. This book covers the end-to-end ML process using Microsoft Azure Machine Learning, including data preparation, performing and logging ML training runs, designing training and deployment pipelines, and managing these pipelines via MLOps. The first section shows you how to set up an Azure Machine Learning workspace; ingest and version datasets; as well as preprocess, label, and enrich these datasets for training. In the next two sections, you'll discover how to enrich and train ML models for embedding, classification, and regression. You'll explore advanced NLP techniques, traditional ML models such as boosted trees, modern deep neural networks, recommendation systems, reinforcement learning, and complex distributed ML training techniques - all using Azure Machine Learning. The last section will teach you how to deploy the trained models as a batch pipeline or real-time scoring service using Docker, Azure Machine Learning clusters, Azure Kubernetes Services, and alternative deployment targets. By the end of this book, you’ll be able to combine all the steps you’ve learned by building an MLOps pipeline.
Table of Contents (23 chapters)
1
Section 1: Introduction to Azure Machine Learning
5
Section 2: Data Ingestion, Preparation, Feature Engineering, and Pipelining
11
Section 3: The Training and Optimization of Machine Learning Models
17
Section 4: Machine Learning Model Deployment and Operations

Summary

Distributed ML is a great approach to scaling out your training infrastructure in order to gain speed in your training process. It is applied in many real-world scenarios and is very easy to use with Horovod and Azure Machine Learning.

Parallel execution is similar to hyperparameter searching, while distributed execution is similar to Bayesian optimization, which we discussed in detail in the previous chapter. Distributed executions need methods to perform communication (such as one-to-one, one-to-many, many-to-one, and many-to-many) and synchronization (such as barrier synchronization) efficiently. These so-called collective algorithms are provided by communication backends (MPI, Gloo, and NCCL) and allow efficient GPU-to-GPU communication.

DL frameworks build higher-level abstractions on top of communication backends to perform model-parallel and data-parallel training. In data-parallel training, we partition the input data to compute multiple independent parts of the...