@@ -30,11 +30,25 @@ approximately 80% actually belong to the positive class.
3030Calibration curves
3131------------------
3232
33- The following plot compares how well the probabilistic predictions of
34- different classifiers are calibrated, using :func: `calibration_curve `.
33+ Calibration curves (also known as reliability diagrams) compare how well the
34+ probabilistic predictions of a binary classifier are calibrated. It plots
35+ the true frequency of the positive label against its predicted probability,
36+ for binned predictions.
3537The x axis represents the average predicted probability in each bin. The
3638y axis is the *fraction of positives *, i.e. the proportion of samples whose
37- class is the positive class (in each bin).
39+ class is the positive class (in each bin). The top calibration curve plot
40+ is created with :func: `CalibrationDisplay.from_estimators `, which uses
41+ :func: `calibration_curve ` to calculate the per bin average predicted
42+ probabilities and fraction of positives.
43+ :func: `CalibrationDisplay.from_estimator `
44+ takes as input a fitted classifier, which is used to calculate the predicted
45+ probabilities. The classifier thus must have :term: `predict_proba ` method. For
46+ the few classifiers that do not have a :term: `predict_proba ` method, it is
47+ possible to use :class: `CalibratedClassifierCV ` to calibrate the classifier
48+ outputs to probabilities.
49+
50+ The bottom histogram gives some insight into the behavior of each classifier
51+ by showing the number of samples in each predicted probability bin.
3852
3953.. figure :: ../auto_examples/calibration/images/sphx_glr_plot_compare_calibration_001.png
4054 :target: ../auto_examples/calibration/plot_compare_calibration.html
@@ -161,6 +175,8 @@ mean a better calibrated model.
161175:class: `CalibratedClassifierCV ` supports the use of two 'calibration'
162176regressors: 'sigmoid' and 'isotonic'.
163177
178+ .. _sigmoid_regressor :
179+
164180Sigmoid
165181^^^^^^^
166182
0 commit comments