sklearn/doc/modules/calibration.rst

313 lines
15 KiB
ReStructuredText
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

.. _calibration:
=======================
Probability calibration
=======================
.. currentmodule:: sklearn.calibration
When performing classification you often want not only to predict the class
label, but also obtain a probability of the respective label. This probability
gives you some kind of confidence on the prediction. Some models can give you
poor estimates of the class probabilities and some even do not support
probability prediction (e.g., some instances of
:class:`~sklearn.linear_model.SGDClassifier`).
The calibration module allows you to better calibrate
the probabilities of a given model, or to add support for probability
prediction.
Well calibrated classifiers are probabilistic classifiers for which the output
of the :term:`predict_proba` method can be directly interpreted as a confidence
level.
For instance, a well calibrated (binary) classifier should classify the samples such
that among the samples to which it gave a :term:`predict_proba` value close to, say,
0.8, approximately 80% actually belong to the positive class.
Before we show how to re-calibrate a classifier, we first need a way to detect how
good a classifier is calibrated.
.. note::
Strictly proper scoring rules for probabilistic predictions like
:func:`sklearn.metrics.brier_score_loss` and
:func:`sklearn.metrics.log_loss` assess calibration (reliability) and
discriminative power (resolution) of a model, as well as the randomness of the data
(uncertainty) at the same time. This follows from the well-known Brier score
decomposition of Murphy [1]_. As it is not clear which term dominates, the score is
of limited use for assessing calibration alone (unless one computes each term of
the decomposition). A lower Brier loss, for instance, does not necessarily
mean a better calibrated model, it could also mean a worse calibrated model with much
more discriminatory power, e.g. using many more features.
.. _calibration_curve:
Calibration curves
------------------
Calibration curves, also referred to as *reliability diagrams* (Wilks 1995 [2]_),
compare how well the probabilistic predictions of a binary classifier are calibrated.
It plots the frequency of the positive label (to be more precise, an estimation of the
*conditional event probability* :math:`P(Y=1|\text{predict_proba})`) on the y-axis
against the predicted probability :term:`predict_proba` of a model on the x-axis.
The tricky part is to get values for the y-axis.
In scikit-learn, this is accomplished by binning the predictions such that the x-axis
represents the average predicted probability in each bin.
The y-axis is then the *fraction of positives* given the predictions of that bin, i.e.
the proportion of samples whose class is the positive class (in each bin).
The top calibration curve plot is created with
:func:`CalibrationDisplay.from_estimator`, which uses :func:`calibration_curve` to
calculate the per bin average predicted probabilities and fraction of positives.
:func:`CalibrationDisplay.from_estimator`
takes as input a fitted classifier, which is used to calculate the predicted
probabilities. The classifier thus must have :term:`predict_proba` method. For
the few classifiers that do not have a :term:`predict_proba` method, it is
possible to use :class:`CalibratedClassifierCV` to calibrate the classifier
outputs to probabilities.
The bottom histogram gives some insight into the behavior of each classifier
by showing the number of samples in each predicted probability bin.
.. figure:: ../auto_examples/calibration/images/sphx_glr_plot_compare_calibration_001.png
:target: ../auto_examples/calibration/plot_compare_calibration.html
:align: center
.. currentmodule:: sklearn.linear_model
:class:`LogisticRegression` is more likely to return well calibrated predictions by itself as it has a
canonical link function for its loss, i.e. the logit-link for the :ref:`log_loss`.
In the unpenalized case, this leads to the so-called **balance property**, see [8]_ and :ref:`Logistic_regression`.
In the plot above, data is generated according to a linear mechanism, which is
consistent with the :class:`LogisticRegression` model (the model is 'well specified'),
and the value of the regularization parameter `C` is tuned to be
appropriate (neither too strong nor too low). As a consequence, this model returns
accurate predictions from its `predict_proba` method.
In contrast to that, the other shown models return biased probabilities; with
different biases per model.
.. currentmodule:: sklearn.naive_bayes
:class:`GaussianNB` (Naive Bayes) tends to push probabilities to 0 or 1 (note the counts
in the histograms). This is mainly because it makes the assumption that
features are conditionally independent given the class, which is not the
case in this dataset which contains 2 redundant features.
.. currentmodule:: sklearn.ensemble
:class:`RandomForestClassifier` shows the opposite behavior: the histograms
show peaks at probabilities approximately 0.2 and 0.9, while probabilities
close to 0 or 1 are very rare. An explanation for this is given by
Niculescu-Mizil and Caruana [3]_: "Methods such as bagging and random
forests that average predictions from a base set of models can have
difficulty making predictions near 0 and 1 because variance in the
underlying base models will bias predictions that should be near zero or one
away from these values. Because predictions are restricted to the interval
[0,1], errors caused by variance tend to be one-sided near zero and one. For
example, if a model should predict p = 0 for a case, the only way bagging
can achieve this is if all bagged trees predict zero. If we add noise to the
trees that bagging is averaging over, this noise will cause some trees to
predict values larger than 0 for this case, thus moving the average
prediction of the bagged ensemble away from 0. We observe this effect most
strongly with random forests because the base-level trees trained with
random forests have relatively high variance due to feature subsetting." As
a result, the calibration curve shows a characteristic sigmoid shape, indicating that
the classifier could trust its "intuition" more and return probabilities closer
to 0 or 1 typically.
.. currentmodule:: sklearn.svm
:class:`LinearSVC` (SVC) shows an even more sigmoid curve than the random forest, which
is typical for maximum-margin methods (compare Niculescu-Mizil and Caruana [3]_), which
focus on difficult to classify samples that are close to the decision boundary (the
support vectors).
Calibrating a classifier
------------------------
.. currentmodule:: sklearn.calibration
Calibrating a classifier consists of fitting a regressor (called a
*calibrator*) that maps the output of the classifier (as given by
:term:`decision_function` or :term:`predict_proba`) to a calibrated probability
in [0, 1]. Denoting the output of the classifier for a given sample by :math:`f_i`,
the calibrator tries to predict the conditional event probability
:math:`P(y_i = 1 | f_i)`.
Ideally, the calibrator is fit on a dataset independent of the training data used to
fit the classifier in the first place.
This is because performance of the classifier on its training data would be
better than for novel data. Using the classifier output of training data
to fit the calibrator would thus result in a biased calibrator that maps to
probabilities closer to 0 and 1 than it should.
Usage
-----
The :class:`CalibratedClassifierCV` class is used to calibrate a classifier.
:class:`CalibratedClassifierCV` uses a cross-validation approach to ensure
unbiased data is always used to fit the calibrator. The data is split into k
`(train_set, test_set)` couples (as determined by `cv`). When `ensemble=True`
(default), the following procedure is repeated independently for each
cross-validation split: a clone of `base_estimator` is first trained on the
train subset. Then its predictions on the test subset are used to fit a
calibrator (either a sigmoid or isotonic regressor). This results in an
ensemble of k `(classifier, calibrator)` couples where each calibrator maps
the output of its corresponding classifier into [0, 1]. Each couple is exposed
in the `calibrated_classifiers_` attribute, where each entry is a calibrated
classifier with a :term:`predict_proba` method that outputs calibrated
probabilities. The output of :term:`predict_proba` for the main
:class:`CalibratedClassifierCV` instance corresponds to the average of the
predicted probabilities of the `k` estimators in the `calibrated_classifiers_`
list. The output of :term:`predict` is the class that has the highest
probability.
When `ensemble=False`, cross-validation is used to obtain 'unbiased'
predictions for all the data, via
:func:`~sklearn.model_selection.cross_val_predict`.
These unbiased predictions are then used to train the calibrator. The attribute
`calibrated_classifiers_` consists of only one `(classifier, calibrator)`
couple where the classifier is the `base_estimator` trained on all the data.
In this case the output of :term:`predict_proba` for
:class:`CalibratedClassifierCV` is the predicted probabilities obtained
from the single `(classifier, calibrator)` couple.
The main advantage of `ensemble=True` is to benefit from the traditional
ensembling effect (similar to :ref:`bagging`). The resulting ensemble should
both be well calibrated and slightly more accurate than with `ensemble=False`.
The main advantage of using `ensemble=False` is computational: it reduces the
overall fit time by training only a single base classifier and calibrator
pair, decreases the final model size and increases prediction speed.
Alternatively an already fitted classifier can be calibrated by setting
`cv="prefit"`. In this case, the data is not split and all of it is used to
fit the regressor. It is up to the user to
make sure that the data used for fitting the classifier is disjoint from the
data used for fitting the regressor.
:class:`CalibratedClassifierCV` supports the use of two regression techniques
for calibration via the `method` parameter: `"sigmoid"` and `"isotonic"`.
.. _sigmoid_regressor:
Sigmoid
^^^^^^^
The sigmoid regressor, `method="sigmoid"` is based on Platt's logistic model [4]_:
.. math::
p(y_i = 1 | f_i) = \frac{1}{1 + \exp(A f_i + B)} \,,
where :math:`y_i` is the true label of sample :math:`i` and :math:`f_i`
is the output of the un-calibrated classifier for sample :math:`i`. :math:`A`
and :math:`B` are real numbers to be determined when fitting the regressor via
maximum likelihood.
The sigmoid method assumes the :ref:`calibration curve <calibration_curve>`
can be corrected by applying a sigmoid function to the raw predictions. This
assumption has been empirically justified in the case of :ref:`svm` with
common kernel functions on various benchmark datasets in section 2.1 of Platt
1999 [4]_ but does not necessarily hold in general. Additionally, the
logistic model works best if the calibration error is symmetrical, meaning
the classifier output for each binary class is normally distributed with
the same variance [7]_. This can be a problem for highly imbalanced
classification problems, where outputs do not have equal variance.
In general this method is most effective for small sample sizes or when the
un-calibrated model is under-confident and has similar calibration errors for both
high and low outputs.
Isotonic
^^^^^^^^
The `method="isotonic"` fits a non-parametric isotonic regressor, which outputs
a step-wise non-decreasing function, see :mod:`sklearn.isotonic`. It minimizes:
.. math::
\sum_{i=1}^{n} (y_i - \hat{f}_i)^2
subject to :math:`\hat{f}_i \geq \hat{f}_j` whenever
:math:`f_i \geq f_j`. :math:`y_i` is the true
label of sample :math:`i` and :math:`\hat{f}_i` is the output of the
calibrated classifier for sample :math:`i` (i.e., the calibrated probability).
This method is more general when compared to 'sigmoid' as the only restriction
is that the mapping function is monotonically increasing. It is thus more
powerful as it can correct any monotonic distortion of the un-calibrated model.
However, it is more prone to overfitting, especially on small datasets [6]_.
Overall, 'isotonic' will perform as well as or better than 'sigmoid' when
there is enough data (greater than ~ 1000 samples) to avoid overfitting [3]_.
.. note:: Impact on ranking metrics like AUC
It is generally expected that calibration does not affect ranking metrics such as
ROC-AUC. However, these metrics might differ after calibration when using
`method="isotonic"` since isotonic regression introduces ties in the predicted
probabilities. This can be seen as within the uncertainty of the model predictions.
In case, you strictly want to keep the ranking and thus AUC scores, use
`method="sigmoid"` which is a strictly monotonic transformation and thus keeps
the ranking.
Multiclass support
^^^^^^^^^^^^^^^^^^
Both isotonic and sigmoid regressors only
support 1-dimensional data (e.g., binary classification output) but are
extended for multiclass classification if the `base_estimator` supports
multiclass predictions. For multiclass predictions,
:class:`CalibratedClassifierCV` calibrates for
each class separately in a :ref:`ovr_classification` fashion [5]_. When
predicting
probabilities, the calibrated probabilities for each class
are predicted separately. As those probabilities do not necessarily sum to
one, a postprocessing is performed to normalize them.
.. rubric:: Examples
* :ref:`sphx_glr_auto_examples_calibration_plot_calibration_curve.py`
* :ref:`sphx_glr_auto_examples_calibration_plot_calibration_multiclass.py`
* :ref:`sphx_glr_auto_examples_calibration_plot_calibration.py`
* :ref:`sphx_glr_auto_examples_calibration_plot_compare_calibration.py`
.. rubric:: References
.. [1] Allan H. Murphy (1973).
:doi:`"A New Vector Partition of the Probability Score"
<10.1175/1520-0450(1973)012%3C0595:ANVPOT%3E2.0.CO;2>`
Journal of Applied Meteorology and Climatology
.. [2] `On the combination of forecast probabilities for
consecutive precipitation periods.
<https://journals.ametsoc.org/waf/article/5/4/640/40179>`_
Wea. Forecasting, 5, 640650., Wilks, D. S., 1990a
.. [3] `Predicting Good Probabilities with Supervised Learning
<https://www.cs.cornell.edu/~alexn/papers/calibration.icml05.crc.rev3.pdf>`_,
A. Niculescu-Mizil & R. Caruana, ICML 2005
.. [4] `Probabilistic Outputs for Support Vector Machines and Comparisons
to Regularized Likelihood Methods.
<https://www.cs.colorado.edu/~mozer/Teaching/syllabi/6622/papers/Platt1999.pdf>`_
J. Platt, (1999)
.. [5] `Transforming Classifier Scores into Accurate Multiclass
Probability Estimates.
<https://dl.acm.org/doi/pdf/10.1145/775047.775151>`_
B. Zadrozny & C. Elkan, (KDD 2002)
.. [6] `Predicting accurate probabilities with a ranking loss.
<https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4180410/>`_
Menon AK, Jiang XJ, Vembu S, Elkan C, Ohno-Machado L.
Proc Int Conf Mach Learn. 2012;2012:703-710
.. [7] `Beyond sigmoids: How to obtain well-calibrated probabilities from
binary classifiers with beta calibration
<https://projecteuclid.org/euclid.ejs/1513306867>`_
Kull, M., Silva Filho, T. M., & Flach, P. (2017).
.. [8] Mario V. Wüthrich, Michael Merz (2023).
:doi:`"Statistical Foundations of Actuarial Learning and its Applications"
<10.1007/978-3-031-12409-9>`
Springer Actuarial