book

Practical Deep Learning at Scale with MLflow

by Yong Liu

July 2022

Intermediate to advanced

288 pages

6h 22m

English

Packt Publishing

Read now

Unlock full access

ForewordContributorsAbout the authorAbout the reviewers
Who this book is forWhat this book coversTo get the most out of this bookDownload the example code filesDownload the color imagesConventions usedGet in touchShare Your Thoughts
Technical requirements Understanding the DL life cycle and MLOps challengesImplementing a basic DL sentiment classifierUnderstanding DL's full life cycle developmentUnderstanding MLOps challengesUnderstanding DL data challengesUnderstanding DL model challengesUnderstanding DL code challengesUnderstanding DL explainability challengesSummaryFurther reading
Technical requirements Setting up MLflowSetting up MLflow locally using minicondaSetting up MLflow to interact with a remote MLflow serverImplementing our first DL experiment with MLflow autologgingExploring MLflow's components and usage patternsExploring experiments and runs in MLflowExploring MLflow models and their usagesExploring MLflow code tracking and its usagesSummaryFurther reading
Technical requirementsSetting up a full-fledged local MLflow tracking serverTracking model provenanceUnderstanding the open provenance tracking frameworkImplementing MLflow model trackingTracking model metrics Tracking model parametersSummaryFurther reading
Technical requirements Tracking notebook and pipeline versioningPipeline trackingTracking locally, privately built Python librariesTracking data versioning in Delta LakeAn example of tracking data using MLflowSummaryFurther reading
Technical requirementsAn overview of different execution scenarios and environmentsRunning locally with local codeRunning remote code in GitHub locallyRunning local code remotely in the cloudRunning remotely in the cloud with remote code in GitHubSummaryFurther reading

Technical requirements Understanding automatic HPO for DL pipelinesTypes of hyperparameters and their challengesHow HPO works and which ones to chooseCreating HPO-ready DL models with Ray Tune and MLflow Setting up Ray Tune and MLflow Creating the Ray Tune trainable for the DL modelCreating the Ray Tune HPO run functionRunning the first Ray Tune HPO experiment with MLflowRunning HPO with Ray Tune using Optuna and HyperBandSummaryFurther reading
Technical requirementsUnderstanding patterns of DL inference pipelinesUnderstanding the MLflow Model Python Function APIImplementing a custom MLflow Python model Implementing preprocessing and postprocessing steps in a DL inference pipelineImplementing language detection preprocessing logicImplementing caching preprocessing and postprocessing logicImplementing response composition postprocessing logicImplementing an inference pipeline as a new entry point in the main MLprojectSummaryFurther reading
Technical requirements Understanding different deployment tools and host environmentsDeploying locally for batch and web service inference Batch inferenceModel as a web serviceDeploying using Ray Serve and MLflow deployment pluginsDeploying to AWS SageMaker – a complete end-to-end guideStep 1: Build a local SageMaker Docker imageStep 2: Add additional model artifacts layers onto the SageMaker Docker imageStep 3: Test local deployment with the newly built SageMaker Docker imageStep 4: Push the SageMaker Docker image to AWS Elastic Container RegistryStep 5: Deploy the inference pipeline model to create a SageMaker endpointStep 6: Query the SageMaker endpoint for online inferenceSummaryFurther reading
Technical requirementsUnderstanding the categories and audience of explainabilityAudience: who needs to knowStage: when to provide an explanation in the DL life cycleScope: which prediction needs explanationInput data format: what is the format of the input dataOutput data format: what is the format of the output explanationProblem type: what is the machine learning problem typeObjectives type: what is the motivation or goal to explainMethod type: what is the specific post-hoc explanation method usedExploring the SHAP Explainability toolboxExploring the Transformers Interpret toolboxSummaryFurther reading
Technical requirementsUnderstanding current MLflow explainability integrationImplementing a SHAP explanation using the MLflow artifact logging APIImplementing a SHAP explainer using the MLflow pyfunc APICreating and logging an MLflow pyfunc explainerDeploying an MLflow pyfunc explainer for an EaaSUsing an MLflow pyfunc explainer for batch explanationSummaryFurther reading
Other Books You May EnjoyPackt is searching for authors like youShare Your Thoughts

Overview

Practical Deep Learning at Scale with MLflow is your comprehensive guide to mastering the full life-cycle process of deep learning. Through step-by-step instructions, you will learn how to bridge the gap between experimentation and production using the MLflow platform. This book emphasizes reproducibility, scalability, and practical deployment of deep learning models.

What this Book will help me do

Master the MLOps practices for managing the deep learning life cycle.
Learn how to utilize MLflow for tracking code, data, and models.
Develop skills to deploy scalable deep learning pipelines.
Implement explainability into deep learning solutions using SHAP.
Gain expertise in hyperparameter optimization using frameworks like Ray Tune.

Author(s)

Yong Liu, the author of this book, is an expert in machine learning and deep learning with a strong focus on scalable solutions. With years of experience in implementing AI frameworks in commercial environments, Yong brings deep insights into practical AI pipeline development. His pedagogical style balances theory with applied best practices to help readers quickly grasp complex concepts.

Who is it for?

This book is designed for professionals such as data scientists, machine learning engineers, and AI practitioners who are looking to scale their expertise in deep learning. It is particularly beneficial for those already familiar with core machine learning concepts and seeking to implement production-capable solutions.