Abstention Zero Coverage Policy Tool

Description

Report coverage and accepted accuracy under a zero-coverage policy.

Classifier metrics count particular outcomes on a declared population and positive-label convention. Precision, recall and accepted-only measures use different denominators. A reporting value at an empty denominator is a policy choice, while one cell, derivative or contribution does not constitute a complete performance evaluation.

Report coverage and accepted accuracy under a zero-coverage policy.

When to use this tool

Use Abstention Zero Coverage Policy when checking the stated confusion-count, metric contribution or abstention-denominator convention.

How to use the tool and how it works

Enter accepted, correct, total, zeroCoveragePolicy. Use the units and conventions below.
Run the tool. Correct any format, range, or cross-field input error before interpreting the output.
Read coverage, accuracyDefined, reportedAccuracy. Inspect accuracyDefined before interpreting the associated values.

Report coverage and accepted accuracy under a zero-coverage policy. Require total >= accepted >= correct >= 0 with positive total; accuracy = correct/accepted when nonempty, otherwise zero or one by policy, with accuracyDefined false only for undefined policy.^[1]

Classifier metrics use different outcome supports and denominator conventions. A zero-denominator reporting policy does not supply missing observed outcomes.^[2]

Input field	Meaning and units	Accepted input
accepted	Accepted	Required. Safe integer; minimum 0.
correct	Correct	Required. Safe integer; minimum 0.
total	Total	Required. Safe integer; minimum 1.
zeroCoveragePolicy	Zero Coverage Policy	Required. Text; one of "zero", "one", "undefined"; maximum UTF-16 length 8192.

Output field	Interpretation	Returned shape
coverage	Coverage	Required. Finite number.
accuracyDefined	At accepted=0, false only for undefined policy; a policy-defined value is not observed accuracy.	Required. Boolean.
reportedAccuracy	Reported Accuracy	Required. Finite number.

json

{
  "input": {
    "accepted": 0,
    "correct": 0,
    "total": 10,
    "zeroCoveragePolicy": "undefined"
  },
  "output": {
    "coverage": 0,
    "accuracyDefined": false,
    "reportedAccuracy": 0
  }
}

Limitations and assumptions

A policy-defined zero/one at empty coverage is a reporting convention rather than observed predictive accuracy.
Inputs must satisfy both the field types and the cross-field conditions above. Invalid input causes an error; a successful false flag or documented placeholder is a different result.
Numeric calculations use binary floating point unless exact decimal-integer arithmetic is explicitly stated. No hidden measurement tolerance or decimal rounding is added; nonfinite or out-of-contract results are rejected. Numeric signed-zero results are exported as zero.

Alternative or Complementary methods

Evaluate the complete confusion table, relevant class supports and abstention coverage together; use suitable uncertainty and validation methods for conclusions about predictive performance.

References

Calculation and input contract: classification/abstention-zero-coverage-policy — tool.io
Model evaluation: classification metrics — scikit-learn developers

Similar or alternative tools

Binary Abstention Coverage Count
Return acceptance coverage from positive, negative and abstained counts.
Classification: Precision Zero Denominator Policy
Return precision with an explicit zero-denominator report policy.