A calibration curve is a diagnostic plot for probabilistic classifiers. It divides predicted probabilities into bins (typically ten equal-width bins from 0 to 1), then for each bin computes the mean predicted probability and the empirical fraction of positive outcomes. The x-axis shows the mean predicted probability and the y-axis shows the observed frequency of positives. A perfectly calibrated model follows the diagonal y=x reference line; deviations above indicate underconfidence, deviations below indicate overconfidence. The curve persists through software libraries (scikit-learn, matplotlib), statistical practice, and machine learning model evaluation pipelines. [formal: calibration_curve | substrate: mind | horizon: a moment | explicit: yes | epoch: 0.01]
Accepted ontology entry
calibration
A calibration curve is a diagnostic plot for probabilistic classifiers. It divides predicted probabilities into bins (typically ten equal-width bins from 0 to 1), then for each bin computes the mean predicted probability and the empirical…
Definition
Why it is in scope
Calibration is the human-made practice of aligning an instrument, measure, or judgment against a known standard. It is a procedure created by metrology and epistemology to reduce systematic error, and persists through institutional protocols, traceable standards, and the disciplined habit of periodic re-verification.
Names and aliases
- calibrationen · CANONICAL
Relations from this entry
- cmr9uz3vv00elhcxfruyltnd4DEPENDS_ON →
Calibration needs measurement to operate now — it cannot calibrate instruments without measuring them against a standard. The dependency is constitutive: remove measurement and calibration ceases to exist.
- cmr9u6cjb00cyhcxf565q5jsrDEPENDS_ON →
Removal test: remove the standard and calibration ceases to operate — calibration IS the comparison of an instrument against a known standard. Without a standard, there is nothing to calibrate against. Law 8 satisfied.
- cmr9uz3vv00elhcxfruyltnd4INSTANCE_OF →
Calibration IS a specific kind of measurement — the process of comparing a measuring instrument against a known standard to verify/adjust its accuracy. A competent speaker would call calibration 'a measurement procedure.'
- cms5rqg1s02mmi9iq606htcu1SERVES →
Calibration is designed and maintained for the sake of precision — its purpose is to adjust an instrument so it produces precise measurements. Law 8d: servant(calibration)→master(precision). The removal test also passes: remove precision and calibration loses its target.
Relations to this entry
- cmsg0ybma00x0qszgfuwzihxg← DEPENDS_ON
Calibration-error measures deviation between predicted and actual values within a calibration framework. Remove calibration as a concept and calibration-error loses all meaning — it cannot operate without the calibration framework to define what 'error' means in context. Removal test passes.
- cmsg6ymgz01d9qszgy5qw5u02← DEPENDS_ON
Probability-calibration is a specific type of calibration practice applied to probabilistic predictions. Remove calibration as a general concept and probability-calibration has no framework to extend. The specific practice depends on the general concept it instantiates. Removal test passes.
- cmsg6ymgz01d9qszgy5qw5u02← DERIVED_FROM
Probability-calibration is a newer, specialized practice derived from the older, more general practice of calibration. Which-came-first test: calibration (epoch 0.19) predates probability-calibration (epoch 0.95). The specific practice extends the general framework.
Record identity
- Created
- Jul 17, 2026, 1:04 PM UTC
- Content hash
- 79c40c09fbef3bdbbe3ff1f987927f27a5421c43d6921d2e2f69e26be8976540