liveBook · Manning

Deployment

As explained in Data Without Labels

Model deployment is the process of making a trained machine-learning model available to process inputs and produce predictions in an operational environment.

Model deployment moves a trained model from development into production, where it processes inputs and produces predictions for real-world business purposes. It turns model outputs into predictions, recommendations, optimizations, and automated workflows. section 11.11 section 11.12

Deployment belongs to an end-to-end process: define the business problem, assess data feasibility, clean and prepare data, perform exploratory data analysis, model and obtain approval, deploy, document, and maintain or refresh the model. section 11.0 section 11.1

Choose a deployment strategy according to data volume, latency, device constraints, and rollout risk:

  • Batch deployment processes accumulated data offline in large groups, such as periodically clustering retail customers. section 11.13
  • Real-time deployment generates low-latency predictions for incoming events, such as classifying simultaneous credit-card transactions. section 11.13
  • Edge deployment runs a small, computationally efficient model on resource-limited devices when connectivity may be unavailable. section 11.13
  • Canary deployment releases a model to a subset of users first, using early feedback before wider release. section 11.13
  • A/B testing compares model versions while serving different user groups, supporting controlled evaluation and release decisions. section 11.19

Operational deployment requires accuracy and resource monitoring, scalability, security and compliance, reproducibility, drift detection, continuous integration and deployment, user feedback, versioning, and rollback. Maintenance continues after release because business conditions and data can change. section 11.14 section 11.19

Fraud detection

A credit-card system can deploy a real-time model that evaluates each incoming transaction with low latency. Monitoring, versioning, and rollback help manage performance and release risk. section 11.13 section 11.14

Figure 11.1 The complete machine learning modeling process
figure

Use the complete modeling-process diagram to navigate deployment alongside business definition, data preparation, modeling, documentation, and maintenance. section 11.1

How Data Without Labels covers it

How deep it goes
Explains the end-to-end stages, deployment strategies, and operational safeguards, then stops short of detailed infrastructure implementation.
Examples
Uses retail customer clustering, credit-card fraud detection, edge devices, canary releases, and A/B testing.
What it assumes
Assumes familiarity with business definition, data discovery, cleaning, exploratory analysis, modeling, and stakeholder approval.
What you can do after
Operate a deployed model through Flask while monitoring performance, drift, security, reproducibility, versioning, and rollback.
How it connects
Connects deployment with data preparation, exploratory data analysis, model maintenance, model drift, continuous deployment, version control, and rollback.
Cover of Data Without Labels
liveBookmanning.com
Read Data Without Labels on liveBook Get the book at manning.com

Also explained in

Compare how 23 books explain deployment →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Data Without Labels. The section links point to the parts of the book it draws on.