Examples
Examples
Try the starting example
- Records: unique ID, finite decision score, binary label
- A,0.9,1 B,0.8,0 C,0.7,1
- Uninterpolated average precision
- 0.833333
- Descending distinct score thresholds
- 0.9, 0.8, 0.7
- Precision at thresholds
- 1, 0.5, 0.666666666667
- Recall at thresholds
- 0.5, 0.5, 1
- Observed positive fraction
- 0.666667
Try another scenario
- Records: unique ID, finite decision score, binary label
- A,0.9,1 B,0.8,0 C,0.85,1
- Uninterpolated average precision
- 1
- Descending distinct score thresholds
- 0.9, 0.85, 0.8
- Precision at thresholds
- 1, 1, 0.666666666667
- Recall at thresholds
- 0.5, 1, 1
- Observed positive fraction
- 0.666667
Calculation method
Calculation method
Uninterpolated average precision
Sum recall increase × precision after each entire tied-score group.
Descending distinct score thresholds
Distinct scores descending.
Precision at thresholds
TP/(TP+FP) after including every tied record.
Recall at thresholds
TP/actual positives.
Observed positive fraction
Actual positives/record count.
Symbols and units
- a
- Records: unique ID, finite decision score, binary label
Use the units shown. A list keeps the order in which you enter its values. The symbol ^ means “raised to the power of”.
Assumptions & limits
Assumptions & limits
- Right-step, noninterpolated average precision; this is not trapezoidal precision-recall area. Tied scores enter together. Inputs are supplied labels and scores; no model is run.
Are these inputs private?
These calculations run in your browser. A shared link contains the values you choose to share. Only questions submitted to the separate AI finder are sent to OpenAI.
Further reading
scikit-learn: average_precision_score ↗
Background reading; this does not imply endorsement or independent review of SumFizz.