Description
Return top-two scores and their relative gap.
Classification scores rank labels or locate a prediction relative to a chosen decision boundary. Raw scores and logits need not be calibrated probabilities; exact ties, score direction and inclusion of boundary values are separate policy choices that can change the reported decision.
Return top-two scores and their relative gap.
When to use this tool
Use Top Two Score Relative Gap when checking the stated score comparison, tie or decision-band policy.
How to use the tool and how it works
- Enter scores. Use the units and conventions below.
- Run the tool. Correct any format, range, or cross-field input error before interpreting the output.
- Read top, second, relativeGap, defined. Inspect defined before interpreting the associated values.
Return top-two scores and their relative gap. Require at least two scores; sort descending, relativeGap = (top-second)/abs(top) when top != 0, otherwise zero with defined false.1
| Input field | Meaning and units | Accepted input |
|---|---|---|
| scores | Scores | Required. Array of finite number; maximum items 256. |
| Output field | Interpretation | Returned shape |
|---|---|---|
| top | Top | Required. Finite number. |
| second | Second | Required. Finite number. |
| relativeGap | (top-second)/abs(top) when top is nonzero; zero placeholder otherwise, even for a negative runner-up. | Required. Finite number. |
| defined | Reporting flag under the explicit denominator or information policy above; inspect any zero placeholder or policy-defined value. | Required. Boolean. |
{
"input": {
"scores": [
2,
5,
3
]
},
"output": {
"top": 5,
"second": 3,
"relativeGap": 0.4,
"defined": true
}
}Limitations and assumptions
- Scores need not be probabilities; a relative gap does not establish calibrated confidence or decision reliability.
- 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
Use a complete evaluation dataset and an explicit decision policy to assess performance; fit and validate calibration separately when a probability interpretation is required.
References
Similar or alternative tools
- Predicted Class Margin Runner Up
Return winning and runner-up score indexes and margin.
- Confusion Row Diagonal Dominance Gap Tool
Calculate confusion row diagonal dominance gap from explicit caller-supplied inputs.