Metrics object

As explained in Designing Deep Learning Systems: A guide for software engineers

A metadata entity that stores quantitative measurements produced during a training run.

A metrics object stores quantitative measurements produced during training, such as a model’s F2 score. It belongs to the metadata schema and links to the training run that produced it. section 8.3.2

Use the training-run link to:

• Trace model performance measurements back to the execution that produced them. section 8.3.2
• Relate metrics to input parameters, model files, experiments, and other run metadata. section 8.3.2

Figure 8.5 An entity relationship diagram of the data schema of the metadata storage

Inspect the entity relationship diagram to see how metrics connect to training runs and related metadata objects. section 8.3.2

How Designing Deep Learning Systems: A guide for software engineers covers it

How deep it goes

Defines the object’s stored measurements and schema relationships, without specifying metric serialization or implementation details.

Examples

Uses a model F2 score as the representative training metric.

What it assumes

Assumes training-run and metadata-schema concepts while establishing the object’s role from those concepts.

What you can do after

Diagnose/debug model performance by tracing measurements from a training run to its inputs and outputs.

How it connects

Connects metrics to training runs, parameters, experiments, models, and metadata stores.

liveBookmanning.com

Read Designing Deep Learning Systems: A guide for software engineers on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Designing Deep Learning Systems: A guide for software engineers. The section links point to the parts of the book it draws on.