Description
Renormalize retained classes after removing an abstention component.
Class prevalence, probability mass and training weights describe different distributions. Removing a class, changing class weights or correcting prior odds changes a particular numerical basis; none of those operations by itself estimates population shift or establishes calibrated predictions.
Renormalize retained classes after removing an abstention component.
When to use this tool
Use Class Probability Renormalization Abstain Term when checking the stated probability-mass, prior-odds or weighting convention.
How to use the tool and how it works
- Enter masses, abstainIndex. Use the units and conventions below.
- Run the tool. Correct any format, range, or cross-field input error before interpreting the output.
- Read removed, factor, retained. Use the output convention below.
Renormalize retained classes after removing an abstention component. Require valid abstainIndex, nonnegative masses with sum within 1e-12 of one and 1-removed > 0; factor = 1/(1-removed), retained multiply by factor.1
| Input field | Meaning and units | Accepted input |
|---|---|---|
| masses | Masses | Required. Array of finite number; maximum items 256. |
| abstainIndex | Abstain Index | Required. Safe integer; minimum 0. |
| Output field | Interpretation | Returned shape |
|---|---|---|
| removed | Removed | Required. Finite number. |
| factor | 1/(1-removed); retained masses are not summed again for the denominator. | Required. Finite number. |
| retained | Retained | Required. Array of finite number; maximum items 256. |
{
"input": {
"masses": [
0.25,
0.25,
0.5
],
"abstainIndex": 2
},
"output": {
"removed": 0.5,
"factor": 2,
"retained": [
0.5,
0.5
]
}
}Limitations and assumptions
- The denominator uses one minus removed, rather than recomputing retained sum; small original roundoff can remain in the result.
- 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
Check the complete class distribution and its sampling basis; use a justified shift model or fitted calibration method when adapting predictions to a new population.
References
Similar or alternative tools
- Classifier Probability Sum Roundoff Gap
Compare total class probability mass with one.
- Multiclass Score Mass Deficit
Return nonnegative score mass and target deficit.