Machine learning system

As explained in Distributed Machine Learning Patterns

A software system that collects data, trains or serves machine learning models, and supports the surrounding operational workflow.

A machine learning system combines data ingestion, model training, model serving, monitoring, and workflow coordination into an operational pipeline. It replaces laptop-scale experiments when datasets, incoming batches, model architectures, serving traffic, or end-to-end workflows exceed a single machine’s capacity. section 1.1.1 section 1.5

Design the system around the flow of data and computation:

• Ingest historical and frequently arriving data for downstream processing. section 1.1.1
• Train and evaluate models using distributed infrastructure when computation outgrows local resources. section 7.1.1
• Serve trained models under potentially heavy prediction traffic, using replicated servers when necessary. section 1.1.1 section 9.0
• Connect components with workflows that support parallel fan-out and fan-in, or asynchronous long-running steps. section 6.0
• Operate the pipeline through scheduling, metadata, observability, monitoring, deployment, and failure-handling practices. section 6.0

End-to-end image-classification system

Download Fashion-MNIST, process it with TensorFlow and Python, train and optimize an image-classification model with Kubeflow on Kubernetes, serve the final model, and compose the components with Argo Workflows. section 7.1.1 section 9.0

Figure 7.14 The architecture diagram of the end-to-end machine learning system we will build

Use the architecture diagram to trace how ingestion, training, serving, and workflow components compose into one scalable end-to-end system. section 7.5

How Distributed Machine Learning Patterns covers it

How deep it goes

Builds from component roles through complete implementation of ingestion, training, serving, and workflow composition.

Examples

Uses Fashion-MNIST, TensorFlow, Python, Kubernetes, Kubeflow, Docker, Argo Workflows, replicated servers, and an end-to-end architecture.

What you can do after

Build an end-to-end distributed pipeline that ingests data, trains models, serves predictions, and coordinates operational steps.

How it connects

Connects data ingestion, model training, model serving, monitoring, workflow orchestration, scheduling, metadata, and distributed infrastructure.

liveBookmanning.com

Read Distributed Machine Learning Patterns on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Machine Learning System DesignA software system that incorporates machine learning models together with the data, infrastructure, and operational components required to produce behavior in an application or production context.
• 

  liveBookmanning.com
  Delivering Machine Learning Projects: From design to deploymentA machine learning system is a deployed software system whose behavior depends on one or more machine learning models and their surrounding data and infrastructure.
• 

  liveBookmanning.com
  Machine Learning Platform EngineeringA software system that incorporates machine learning models and the data, computation, and operational components required to deliver their outputs.

Compare how 4 books explain machine learning system →

Related concepts

• batchingused by · Batching is the grouping of multiple data items, requests, or operations so they can be processed together as a unit.
• data assetuses · A data asset is a collection of data maintained or controlled as a resource for analysis, model development, or system operation.
• data sourceuses · A data source is an origin, system, repository, or process from which data is obtained for use or processing.
• fraud detectionincludes · Fraud detection is the process of identifying transactions, activities, or other events suspected of involving intentional deception or unauthorized gain.
• demand forecastused by · A demand forecast is an estimate or prediction of the quantity of a product, service, or resource that will be needed during a specified future period.
• data ingestionhas part · Data ingestion is the process of acquiring or receiving data from one or more sources and bringing it into a system for storage or processing.

This explainer was generated by Manning's knowledge graph from the text of Distributed Machine Learning Patterns. The section links point to the parts of the book it draws on.