Skip to main content

Data volume

Data volume asks how much data you have available for the problem you want to solve.

This is not an exact number. What counts as small, medium, or big depends on many factors especially the data modality.

You should choose the option that best matches your practical situation.

Options in this tool​

Small​

You have relatively limited data for the task you want to solve.
This often means:

  • only a small number of labelled samples,
  • expensive or difficult data collection,
  • limited coverage of rare cases or edge cases.

Examples: expert-labelled medical images, field campaigns, rare-event data, small legal datasets.

Medium​

You have enough data to train useful models, but not so much that scale alone solves the problem.
This is often the most common practical setting.

Examples: a moderate annotated image dataset, a few years of operational time series, a reasonably sized tabular dataset from one institution or project.

Big​

You have a large amount of data relative to the problem and modality.
This may include:

  • many labelled samples,
  • high-dimensional data where each sample is large or complex,
  • broad coverage of conditions,
  • enough scale to support larger models or extensive training.

Examples: large text corpora, web-scale image datasets, long-running sensor archives, large operational business datasets, high-resolution simulations or 3D data with few but very large samples.

Why it depends on modality​

The same number of samples can mean very different things depending on the data type.

For example:

  • 10,000 images may be modest for computer vision,
  • 10,000 3D volumes may already be substantial,
  • 10,000 text documents may still be small for language modelling,
  • 10,000 tabular rows may be enough for some structured-data tasks but not for others.