Dataset size

As explained in Machine Learning System Design

The quantity of records or examples contained in a dataset or used for model training.

Dataset size is a practical modeling choice, not simply a count of available records. A sample-wise learning curve shows whether adding training examples still improves validation performance or has reached a plateau. section 9.1 section 9.4.1

Use dataset-size analysis to make decisions about relevance, cost, and product impact:

• Test whether all available history is needed, or whether only the most recent months remain relevant. section 9.4.1
• Measure whether downsampling records reduces computation without noticeably degrading validation metrics. section 9.4.1
• Choose the data granularity deliberately: daily records may capture detail, while weekly records can reduce computation substantially. section 9.4.1
• Treat granularity as a product decision as well as a modeling decision, because it can change service speed and available use cases. section 9.4.1

Supermegaretail demand forecasting

A forecasting team asks whether it needs every day–store–item record. A sample-wise learning curve tests how many samples are needed before validation error plateaus; the team can then compare full data with 20%, 10%, or 5% samples. It also compares daily data with weekly store–item data, which reduces computations by a factor of seven while potentially preserving or improving performance. section 9.4.1

Figure 9.1 Basic representation of a learning curve

Use the learning-curve figure to read how a chosen metric changes as a numerical property such as dataset size changes. section 9.1

How Machine Learning System Design covers it

How deep it goes

Explains dataset-size learning curves and validation plateaus, then applies them to sampling and temporal granularity decisions.

Examples

Compares full, downsampled, daily, and weekly day–store–item datasets in a demand-forecasting scenario.

What it assumes

Assumes familiarity with training data, validation metrics, model hyperparameters, and overfitting versus underfitting.

What you can do after

Diagnose whether more samples are worthwhile and choose a data resolution that balances validation performance, computation, and product needs.

How it connects

Connects dataset size with learning curves, validation error, model complexity, downsampling, demand forecasting, and stock management systems.

liveBookmanning.com

Read Machine Learning System Design on liveBook Get the book at manning.com

Related concepts

• scaling lawused by · A scaling law is an empirical relationship describing how a system’s properties change as its scale or resource levels vary.
• memory-to-data ratiomeasured by · The memory-to-data ratio compares available system memory capacity with the capacity of the data or persistent storage being considered.
• accuracyproduces · Accuracy is the degree to which an output, prediction, or measurement agrees with a correct or reference outcome under a specified criterion.
• parameteraddresses · A model parameter is a numerical value, such as a weight or bias, that is learned or set to determine how a model computes outputs.

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