Perpoint

Perpoint is a term used in data processing and analytics to describe a method or metric that attributes cost, performance, or value to each data point in a dataset. The concept is not a formal standard, but it appears in research papers and industry discussions as a way to reason about scalability, efficiency, and pricing when data points vary in complexity or value.

The concept can refer to several related ideas, including resource per data point (such as energy or compute time per point), throughput per point (points processed per second), or price per data point in data markets or service contracts. It complements aggregate metrics by offering a fine-grained view that supports comparison across pipelines, algorithms, or datasets.

Calculation: perpoint value is typically computed by dividing total resources, such as CPU time, memory, energy, or cost, by the number of data points processed or labeled. For example, if a cleaning job processes 2 million points in 90 minutes using 30 kWh, the perpoint energy is 30 kWh / 2,000,000 = 0.000015 kWh per point.

Applications: In data annotation and labeling pipelines, perpoint rates help managers estimate effort and compensation. In ML and data processing, they support benchmarking and hardware planning, enabling comparison of different algorithms on equal data-point bases. In data marketplaces, perpoint pricing connects data value to a specific data item.

Limitations: The usefulness of perpoint depends on uniformity of data points; points with varying complexity can distort comparisons. It should be used alongside other metrics such as per-batch, per-task, or per-feature measures. The term remains informal and may be interpreted differently in different domains.

See also: per-sample, per-point gradient, per-point pricing, point-based rendering.