.. _svm:
=======================
Support Vector Machines
=======================
.. TODO: Describe tol parameter
.. TODO: Describe max_iter parameter
.. currentmodule:: sklearn.svm
**Support vector machines (SVMs)** are a set of supervised learning
methods used for :ref:`classification `,
:ref:`regression ` and :ref:`outliers detection
`.
The advantages of support vector machines are:
- Effective in high dimensional spaces.
- Still effective in cases where number of dimensions is greater
than the number of samples.
- Uses a subset of training points in the decision function (called
support vectors), so it is also memory efficient.
- Versatile: different :ref:`svm_kernels` can be
specified for the decision function. Common kernels are
provided, but it is also possible to specify custom kernels.
The disadvantages of support vector machines include:
- If the number of features is much greater than the number of
samples, avoid over-fitting in choosing :ref:`svm_kernels` and regularization
term is crucial.
- SVMs do not directly provide probability estimates, these are
calculated using an expensive five-fold cross-validation
(see :ref:`Scores and probabilities `, below).
The support vector machines in scikit-learn support both dense
(``numpy.ndarray`` and convertible to that by ``numpy.asarray``) and
sparse (any ``scipy.sparse``) sample vectors as input. However, to use
an SVM to make predictions for sparse data, it must have been fit on such
data. For optimal performance, use C-ordered ``numpy.ndarray`` (dense) or
``scipy.sparse.csr_matrix`` (sparse) with ``dtype=float64``.
.. _svm_classification:
Classification
==============
:class:`SVC`, :class:`NuSVC` and :class:`LinearSVC` are classes
capable of performing binary and multi-class classification on a dataset.
.. figure:: ../auto_examples/svm/images/sphx_glr_plot_iris_svc_001.png
:target: ../auto_examples/svm/plot_iris_svc.html
:align: center
:class:`SVC` and :class:`NuSVC` are similar methods, but accept slightly
different sets of parameters and have different mathematical formulations (see
section :ref:`svm_mathematical_formulation`). On the other hand,
:class:`LinearSVC` is another (faster) implementation of Support Vector
Classification for the case of a linear kernel. It also
lacks some of the attributes of :class:`SVC` and :class:`NuSVC`, like
`support_`. :class:`LinearSVC` uses `squared_hinge` loss and due to its
implementation in `liblinear` it also regularizes the intercept, if considered.
This effect can however be reduced by carefully fine tuning its
`intercept_scaling` parameter, which allows the intercept term to have a
different regularization behavior compared to the other features. The
classification results and score can therefore differ from the other two
classifiers.
As other classifiers, :class:`SVC`, :class:`NuSVC` and
:class:`LinearSVC` take as input two arrays: an array `X` of shape
`(n_samples, n_features)` holding the training samples, and an array `y` of
class labels (strings or integers), of shape `(n_samples)`::
>>> from sklearn import svm
>>> X = [[0, 0], [1, 1]]
>>> y = [0, 1]
>>> clf = svm.SVC()
>>> clf.fit(X, y)
SVC()
After being fitted, the model can then be used to predict new values::
>>> clf.predict([[2., 2.]])
array([1])
SVMs decision function (detailed in the :ref:`svm_mathematical_formulation`)
depends on some subset of the training data, called the support vectors. Some
properties of these support vectors can be found in attributes
``support_vectors_``, ``support_`` and ``n_support_``::
>>> # get support vectors
>>> clf.support_vectors_
array([[0., 0.],
[1., 1.]])
>>> # get indices of support vectors
>>> clf.support_
array([0, 1]...)
>>> # get number of support vectors for each class
>>> clf.n_support_
array([1, 1]...)
.. topic:: Examples:
* :ref:`sphx_glr_auto_examples_svm_plot_separating_hyperplane.py`,
* :ref:`sphx_glr_auto_examples_svm_plot_svm_nonlinear.py`
* :ref:`sphx_glr_auto_examples_svm_plot_svm_anova.py`,
.. _svm_multi_class:
Multi-class classification
--------------------------
:class:`SVC` and :class:`NuSVC` implement the "one-versus-one"
approach for multi-class classification. In total,
``n_classes * (n_classes - 1) / 2``
classifiers are constructed and each one trains data from two classes.
To provide a consistent interface with other classifiers, the
``decision_function_shape`` option allows to monotonically transform the
results of the "one-versus-one" classifiers to a "one-vs-rest" decision
function of shape ``(n_samples, n_classes)``.
>>> X = [[0], [1], [2], [3]]
>>> Y = [0, 1, 2, 3]
>>> clf = svm.SVC(decision_function_shape='ovo')
>>> clf.fit(X, Y)
SVC(decision_function_shape='ovo')
>>> dec = clf.decision_function([[1]])
>>> dec.shape[1] # 4 classes: 4*3/2 = 6
6
>>> clf.decision_function_shape = "ovr"
>>> dec = clf.decision_function([[1]])
>>> dec.shape[1] # 4 classes
4
On the other hand, :class:`LinearSVC` implements "one-vs-the-rest"
multi-class strategy, thus training `n_classes` models.
>>> lin_clf = svm.LinearSVC(dual="auto")
>>> lin_clf.fit(X, Y)
LinearSVC(dual='auto')
>>> dec = lin_clf.decision_function([[1]])
>>> dec.shape[1]
4
See :ref:`svm_mathematical_formulation` for a complete description of
the decision function.
|details-start|
**Details on multi-class strategies**
|details-split|
Note that the :class:`LinearSVC` also implements an alternative multi-class
strategy, the so-called multi-class SVM formulated by Crammer and Singer
[#8]_, by using the option ``multi_class='crammer_singer'``. In practice,
one-vs-rest classification is usually preferred, since the results are mostly
similar, but the runtime is significantly less.
For "one-vs-rest" :class:`LinearSVC` the attributes ``coef_`` and ``intercept_``
have the shape ``(n_classes, n_features)`` and ``(n_classes,)`` respectively.
Each row of the coefficients corresponds to one of the ``n_classes``
"one-vs-rest" classifiers and similar for the intercepts, in the
order of the "one" class.
In the case of "one-vs-one" :class:`SVC` and :class:`NuSVC`, the layout of
the attributes is a little more involved. In the case of a linear
kernel, the attributes ``coef_`` and ``intercept_`` have the shape
``(n_classes * (n_classes - 1) / 2, n_features)`` and ``(n_classes *
(n_classes - 1) / 2)`` respectively. This is similar to the layout for
:class:`LinearSVC` described above, with each row now corresponding
to a binary classifier. The order for classes
0 to n is "0 vs 1", "0 vs 2" , ... "0 vs n", "1 vs 2", "1 vs 3", "1 vs n", . .
. "n-1 vs n".
The shape of ``dual_coef_`` is ``(n_classes-1, n_SV)`` with
a somewhat hard to grasp layout.
The columns correspond to the support vectors involved in any
of the ``n_classes * (n_classes - 1) / 2`` "one-vs-one" classifiers.
Each support vector ``v`` has a dual coefficient in each of the
``n_classes - 1`` classifiers comparing the class of ``v`` against another class.
Note that some, but not all, of these dual coefficients, may be zero.
The ``n_classes - 1`` entries in each column are these dual coefficients,
ordered by the opposing class.
This might be clearer with an example: consider a three class problem with
class 0 having three support vectors
:math:`v^{0}_0, v^{1}_0, v^{2}_0` and class 1 and 2 having two support vectors
:math:`v^{0}_1, v^{1}_1` and :math:`v^{0}_2, v^{1}_2` respectively. For each
support vector :math:`v^{j}_i`, there are two dual coefficients. Let's call
the coefficient of support vector :math:`v^{j}_i` in the classifier between
classes :math:`i` and :math:`k` :math:`\alpha^{j}_{i,k}`.
Then ``dual_coef_`` looks like this:
+------------------------+------------------------+------------------------+------------------------+------------------------+------------------------+------------------------+
|:math:`\alpha^{0}_{0,1}`|:math:`\alpha^{1}_{0,1}`|:math:`\alpha^{2}_{0,1}`|:math:`\alpha^{0}_{1,0}`|:math:`\alpha^{1}_{1,0}`|:math:`\alpha^{0}_{2,0}`|:math:`\alpha^{1}_{2,0}`|
+------------------------+------------------------+------------------------+------------------------+------------------------+------------------------+------------------------+
|:math:`\alpha^{0}_{0,2}`|:math:`\alpha^{1}_{0,2}`|:math:`\alpha^{2}_{0,2}`|:math:`\alpha^{0}_{1,2}`|:math:`\alpha^{1}_{1,2}`|:math:`\alpha^{0}_{2,1}`|:math:`\alpha^{1}_{2,1}`|
+------------------------+------------------------+------------------------+------------------------+------------------------+------------------------+------------------------+
|Coefficients |Coefficients |Coefficients |
|for SVs of class 0 |for SVs of class 1 |for SVs of class 2 |
+--------------------------------------------------------------------------+-------------------------------------------------+-------------------------------------------------+
|details-end|
.. topic:: Examples:
* :ref:`sphx_glr_auto_examples_svm_plot_iris_svc.py`,
.. _scores_probabilities:
Scores and probabilities
------------------------
The ``decision_function`` method of :class:`SVC` and :class:`NuSVC` gives
per-class scores for each sample (or a single score per sample in the binary
case). When the constructor option ``probability`` is set to ``True``,
class membership probability estimates (from the methods ``predict_proba`` and
``predict_log_proba``) are enabled. In the binary case, the probabilities are
calibrated using Platt scaling [#1]_: logistic regression on the SVM's scores,
fit by an additional cross-validation on the training data.
In the multiclass case, this is extended as per [#2]_.
.. note::
The same probability calibration procedure is available for all estimators
via the :class:`~sklearn.calibration.CalibratedClassifierCV` (see
:ref:`calibration`). In the case of :class:`SVC` and :class:`NuSVC`, this
procedure is builtin in `libsvm`_ which is used under the hood, so it does
not rely on scikit-learn's
:class:`~sklearn.calibration.CalibratedClassifierCV`.
The cross-validation involved in Platt scaling
is an expensive operation for large datasets.
In addition, the probability estimates may be inconsistent with the scores:
- the "argmax" of the scores may not be the argmax of the probabilities
- in binary classification, a sample may be labeled by ``predict`` as
belonging to the positive class even if the output of `predict_proba` is
less than 0.5; and similarly, it could be labeled as negative even if the
output of `predict_proba` is more than 0.5.
Platt's method is also known to have theoretical issues.
If confidence scores are required, but these do not have to be probabilities,
then it is advisable to set ``probability=False``
and use ``decision_function`` instead of ``predict_proba``.
Please note that when ``decision_function_shape='ovr'`` and ``n_classes > 2``,
unlike ``decision_function``, the ``predict`` method does not try to break ties
by default. You can set ``break_ties=True`` for the output of ``predict`` to be
the same as ``np.argmax(clf.decision_function(...), axis=1)``, otherwise the
first class among the tied classes will always be returned; but have in mind
that it comes with a computational cost. See
:ref:`sphx_glr_auto_examples_svm_plot_svm_tie_breaking.py` for an example on
tie breaking.
Unbalanced problems
--------------------
In problems where it is desired to give more importance to certain
classes or certain individual samples, the parameters ``class_weight`` and
``sample_weight`` can be used.
:class:`SVC` (but not :class:`NuSVC`) implements the parameter
``class_weight`` in the ``fit`` method. It's a dictionary of the form
``{class_label : value}``, where value is a floating point number > 0
that sets the parameter ``C`` of class ``class_label`` to ``C * value``.
The figure below illustrates the decision boundary of an unbalanced problem,
with and without weight correction.
.. figure:: ../auto_examples/svm/images/sphx_glr_plot_separating_hyperplane_unbalanced_001.png
:target: ../auto_examples/svm/plot_separating_hyperplane_unbalanced.html
:align: center
:scale: 75
:class:`SVC`, :class:`NuSVC`, :class:`SVR`, :class:`NuSVR`, :class:`LinearSVC`,
:class:`LinearSVR` and :class:`OneClassSVM` implement also weights for
individual samples in the `fit` method through the ``sample_weight`` parameter.
Similar to ``class_weight``, this sets the parameter ``C`` for the i-th
example to ``C * sample_weight[i]``, which will encourage the classifier to
get these samples right. The figure below illustrates the effect of sample
weighting on the decision boundary. The size of the circles is proportional
to the sample weights:
.. figure:: ../auto_examples/svm/images/sphx_glr_plot_weighted_samples_001.png
:target: ../auto_examples/svm/plot_weighted_samples.html
:align: center
:scale: 75
.. topic:: Examples:
* :ref:`sphx_glr_auto_examples_svm_plot_separating_hyperplane_unbalanced.py`
* :ref:`sphx_glr_auto_examples_svm_plot_weighted_samples.py`,
.. _svm_regression:
Regression
==========
The method of Support Vector Classification can be extended to solve
regression problems. This method is called Support Vector Regression.
The model produced by support vector classification (as described
above) depends only on a subset of the training data, because the cost
function for building the model does not care about training points
that lie beyond the margin. Analogously, the model produced by Support
Vector Regression depends only on a subset of the training data,
because the cost function ignores samples whose prediction is close to their
target.
There are three different implementations of Support Vector Regression:
:class:`SVR`, :class:`NuSVR` and :class:`LinearSVR`. :class:`LinearSVR`
provides a faster implementation than :class:`SVR` but only considers the
linear kernel, while :class:`NuSVR` implements a slightly different formulation
than :class:`SVR` and :class:`LinearSVR`. Due to its implementation in
`liblinear` :class:`LinearSVR` also regularizes the intercept, if considered.
This effect can however be reduced by carefully fine tuning its
`intercept_scaling` parameter, which allows the intercept term to have a
different regularization behavior compared to the other features. The
classification results and score can therefore differ from the other two
classifiers. See :ref:`svm_implementation_details` for further details.
As with classification classes, the fit method will take as
argument vectors X, y, only that in this case y is expected to have
floating point values instead of integer values::
>>> from sklearn import svm
>>> X = [[0, 0], [2, 2]]
>>> y = [0.5, 2.5]
>>> regr = svm.SVR()
>>> regr.fit(X, y)
SVR()
>>> regr.predict([[1, 1]])
array([1.5])
.. topic:: Examples:
* :ref:`sphx_glr_auto_examples_svm_plot_svm_regression.py`
.. _svm_outlier_detection:
Density estimation, novelty detection
=======================================
The class :class:`OneClassSVM` implements a One-Class SVM which is used in
outlier detection.
See :ref:`outlier_detection` for the description and usage of OneClassSVM.
Complexity
==========
Support Vector Machines are powerful tools, but their compute and
storage requirements increase rapidly with the number of training
vectors. The core of an SVM is a quadratic programming problem (QP),
separating support vectors from the rest of the training data. The QP
solver used by the `libsvm`_-based implementation scales between
:math:`O(n_{features} \times n_{samples}^2)` and
:math:`O(n_{features} \times n_{samples}^3)` depending on how efficiently
the `libsvm`_ cache is used in practice (dataset dependent). If the data
is very sparse :math:`n_{features}` should be replaced by the average number
of non-zero features in a sample vector.
For the linear case, the algorithm used in
:class:`LinearSVC` by the `liblinear`_ implementation is much more
efficient than its `libsvm`_-based :class:`SVC` counterpart and can
scale almost linearly to millions of samples and/or features.
Tips on Practical Use
=====================
* **Avoiding data copy**: For :class:`SVC`, :class:`SVR`, :class:`NuSVC` and
:class:`NuSVR`, if the data passed to certain methods is not C-ordered
contiguous and double precision, it will be copied before calling the
underlying C implementation. You can check whether a given numpy array is
C-contiguous by inspecting its ``flags`` attribute.
For :class:`LinearSVC` (and :class:`LogisticRegression
`) any input passed as a numpy
array will be copied and converted to the `liblinear`_ internal sparse data
representation (double precision floats and int32 indices of non-zero
components). If you want to fit a large-scale linear classifier without
copying a dense numpy C-contiguous double precision array as input, we
suggest to use the :class:`SGDClassifier
` class instead. The objective
function can be configured to be almost the same as the :class:`LinearSVC`
model.
* **Kernel cache size**: For :class:`SVC`, :class:`SVR`, :class:`NuSVC` and
:class:`NuSVR`, the size of the kernel cache has a strong impact on run
times for larger problems. If you have enough RAM available, it is
recommended to set ``cache_size`` to a higher value than the default of
200(MB), such as 500(MB) or 1000(MB).
* **Setting C**: ``C`` is ``1`` by default and it's a reasonable default
choice. If you have a lot of noisy observations you should decrease it:
decreasing C corresponds to more regularization.
:class:`LinearSVC` and :class:`LinearSVR` are less sensitive to ``C`` when
it becomes large, and prediction results stop improving after a certain
threshold. Meanwhile, larger ``C`` values will take more time to train,
sometimes up to 10 times longer, as shown in [#3]_.
* Support Vector Machine algorithms are not scale invariant, so **it
is highly recommended to scale your data**. For example, scale each
attribute on the input vector X to [0,1] or [-1,+1], or standardize it
to have mean 0 and variance 1. Note that the *same* scaling must be
applied to the test vector to obtain meaningful results. This can be done
easily by using a :class:`~sklearn.pipeline.Pipeline`::
>>> from sklearn.pipeline import make_pipeline
>>> from sklearn.preprocessing import StandardScaler
>>> from sklearn.svm import SVC
>>> clf = make_pipeline(StandardScaler(), SVC())
See section :ref:`preprocessing` for more details on scaling and
normalization.
.. _shrinking_svm:
* Regarding the `shrinking` parameter, quoting [#4]_: *We found that if the
number of iterations is large, then shrinking can shorten the training
time. However, if we loosely solve the optimization problem (e.g., by
using a large stopping tolerance), the code without using shrinking may
be much faster*
* Parameter ``nu`` in :class:`NuSVC`/:class:`OneClassSVM`/:class:`NuSVR`
approximates the fraction of training errors and support vectors.
* In :class:`SVC`, if the data is unbalanced (e.g. many
positive and few negative), set ``class_weight='balanced'`` and/or try
different penalty parameters ``C``.
* **Randomness of the underlying implementations**: The underlying
implementations of :class:`SVC` and :class:`NuSVC` use a random number
generator only to shuffle the data for probability estimation (when
``probability`` is set to ``True``). This randomness can be controlled
with the ``random_state`` parameter. If ``probability`` is set to ``False``
these estimators are not random and ``random_state`` has no effect on the
results. The underlying :class:`OneClassSVM` implementation is similar to
the ones of :class:`SVC` and :class:`NuSVC`. As no probability estimation
is provided for :class:`OneClassSVM`, it is not random.
The underlying :class:`LinearSVC` implementation uses a random number
generator to select features when fitting the model with a dual coordinate
descent (i.e. when ``dual`` is set to ``True``). It is thus not uncommon
to have slightly different results for the same input data. If that
happens, try with a smaller `tol` parameter. This randomness can also be
controlled with the ``random_state`` parameter. When ``dual`` is
set to ``False`` the underlying implementation of :class:`LinearSVC` is
not random and ``random_state`` has no effect on the results.
* Using L1 penalization as provided by ``LinearSVC(penalty='l1',
dual=False)`` yields a sparse solution, i.e. only a subset of feature
weights is different from zero and contribute to the decision function.
Increasing ``C`` yields a more complex model (more features are selected).
The ``C`` value that yields a "null" model (all weights equal to zero) can
be calculated using :func:`l1_min_c`.
.. _svm_kernels:
Kernel functions
================
The *kernel function* can be any of the following:
* linear: :math:`\langle x, x'\rangle`.
* polynomial: :math:`(\gamma \langle x, x'\rangle + r)^d`, where
:math:`d` is specified by parameter ``degree``, :math:`r` by ``coef0``.
* rbf: :math:`\exp(-\gamma \|x-x'\|^2)`, where :math:`\gamma` is
specified by parameter ``gamma``, must be greater than 0.
* sigmoid :math:`\tanh(\gamma \langle x,x'\rangle + r)`,
where :math:`r` is specified by ``coef0``.
Different kernels are specified by the `kernel` parameter::
>>> linear_svc = svm.SVC(kernel='linear')
>>> linear_svc.kernel
'linear'
>>> rbf_svc = svm.SVC(kernel='rbf')
>>> rbf_svc.kernel
'rbf'
See also :ref:`kernel_approximation` for a solution to use RBF kernels that is much faster and more scalable.
Parameters of the RBF Kernel
----------------------------
When training an SVM with the *Radial Basis Function* (RBF) kernel, two
parameters must be considered: ``C`` and ``gamma``. The parameter ``C``,
common to all SVM kernels, trades off misclassification of training examples
against simplicity of the decision surface. A low ``C`` makes the decision
surface smooth, while a high ``C`` aims at classifying all training examples
correctly. ``gamma`` defines how much influence a single training example has.
The larger ``gamma`` is, the closer other examples must be to be affected.
Proper choice of ``C`` and ``gamma`` is critical to the SVM's performance. One
is advised to use :class:`~sklearn.model_selection.GridSearchCV` with
``C`` and ``gamma`` spaced exponentially far apart to choose good values.
.. topic:: Examples:
* :ref:`sphx_glr_auto_examples_svm_plot_rbf_parameters.py`
* :ref:`sphx_glr_auto_examples_svm_plot_svm_nonlinear.py`
Custom Kernels
--------------
You can define your own kernels by either giving the kernel as a
python function or by precomputing the Gram matrix.
Classifiers with custom kernels behave the same way as any other
classifiers, except that:
* Field ``support_vectors_`` is now empty, only indices of support
vectors are stored in ``support_``
* A reference (and not a copy) of the first argument in the ``fit()``
method is stored for future reference. If that array changes between the
use of ``fit()`` and ``predict()`` you will have unexpected results.
|details-start|
**Using Python functions as kernels**
|details-split|
You can use your own defined kernels by passing a function to the
``kernel`` parameter.
Your kernel must take as arguments two matrices of shape
``(n_samples_1, n_features)``, ``(n_samples_2, n_features)``
and return a kernel matrix of shape ``(n_samples_1, n_samples_2)``.
The following code defines a linear kernel and creates a classifier
instance that will use that kernel::
>>> import numpy as np
>>> from sklearn import svm
>>> def my_kernel(X, Y):
... return np.dot(X, Y.T)
...
>>> clf = svm.SVC(kernel=my_kernel)
|details-end|
|details-start|
**Using the Gram matrix**
|details-split|
You can pass pre-computed kernels by using the ``kernel='precomputed'``
option. You should then pass Gram matrix instead of X to the `fit` and
`predict` methods. The kernel values between *all* training vectors and the
test vectors must be provided:
>>> import numpy as np
>>> from sklearn.datasets import make_classification
>>> from sklearn.model_selection import train_test_split
>>> from sklearn import svm
>>> X, y = make_classification(n_samples=10, random_state=0)
>>> X_train , X_test , y_train, y_test = train_test_split(X, y, random_state=0)
>>> clf = svm.SVC(kernel='precomputed')
>>> # linear kernel computation
>>> gram_train = np.dot(X_train, X_train.T)
>>> clf.fit(gram_train, y_train)
SVC(kernel='precomputed')
>>> # predict on training examples
>>> gram_test = np.dot(X_test, X_train.T)
>>> clf.predict(gram_test)
array([0, 1, 0])
|details-end|
.. topic:: Examples:
* :ref:`sphx_glr_auto_examples_svm_plot_custom_kernel.py`.
.. _svm_mathematical_formulation:
Mathematical formulation
========================
A support vector machine constructs a hyper-plane or set of hyper-planes in a
high or infinite dimensional space, which can be used for
classification, regression or other tasks. Intuitively, a good
separation is achieved by the hyper-plane that has the largest distance
to the nearest training data points of any class (so-called functional
margin), since in general the larger the margin the lower the
generalization error of the classifier. The figure below shows the decision
function for a linearly separable problem, with three samples on the
margin boundaries, called "support vectors":
.. figure:: ../auto_examples/svm/images/sphx_glr_plot_separating_hyperplane_001.png
:align: center
:scale: 75
In general, when the problem isn't linearly separable, the support vectors
are the samples *within* the margin boundaries.
We recommend [#5]_ and [#6]_ as good references for the theory and
practicalities of SVMs.
SVC
---
Given training vectors :math:`x_i \in \mathbb{R}^p`, i=1,..., n, in two classes, and a
vector :math:`y \in \{1, -1\}^n`, our goal is to find :math:`w \in
\mathbb{R}^p` and :math:`b \in \mathbb{R}` such that the prediction given by
:math:`\text{sign} (w^T\phi(x) + b)` is correct for most samples.
SVC solves the following primal problem:
.. math::
\min_ {w, b, \zeta} \frac{1}{2} w^T w + C \sum_{i=1}^{n} \zeta_i
\textrm {subject to } & y_i (w^T \phi (x_i) + b) \geq 1 - \zeta_i,\\
& \zeta_i \geq 0, i=1, ..., n
Intuitively, we're trying to maximize the margin (by minimizing
:math:`||w||^2 = w^Tw`), while incurring a penalty when a sample is
misclassified or within the margin boundary. Ideally, the value :math:`y_i
(w^T \phi (x_i) + b)` would be :math:`\geq 1` for all samples, which
indicates a perfect prediction. But problems are usually not always perfectly
separable with a hyperplane, so we allow some samples to be at a distance :math:`\zeta_i` from
their correct margin boundary. The penalty term `C` controls the strength of
this penalty, and as a result, acts as an inverse regularization parameter
(see note below).
The dual problem to the primal is
.. math::
\min_{\alpha} \frac{1}{2} \alpha^T Q \alpha - e^T \alpha
\textrm {subject to } & y^T \alpha = 0\\
& 0 \leq \alpha_i \leq C, i=1, ..., n
where :math:`e` is the vector of all ones,
and :math:`Q` is an :math:`n` by :math:`n` positive semidefinite matrix,
:math:`Q_{ij} \equiv y_i y_j K(x_i, x_j)`, where :math:`K(x_i, x_j) = \phi (x_i)^T \phi (x_j)`
is the kernel. The terms :math:`\alpha_i` are called the dual coefficients,
and they are upper-bounded by :math:`C`.
This dual representation highlights the fact that training vectors are
implicitly mapped into a higher (maybe infinite)
dimensional space by the function :math:`\phi`: see `kernel trick
`_.
Once the optimization problem is solved, the output of
:term:`decision_function` for a given sample :math:`x` becomes:
.. math:: \sum_{i\in SV} y_i \alpha_i K(x_i, x) + b,
and the predicted class correspond to its sign. We only need to sum over the
support vectors (i.e. the samples that lie within the margin) because the
dual coefficients :math:`\alpha_i` are zero for the other samples.
These parameters can be accessed through the attributes ``dual_coef_``
which holds the product :math:`y_i \alpha_i`, ``support_vectors_`` which
holds the support vectors, and ``intercept_`` which holds the independent
term :math:`b`
.. note::
While SVM models derived from `libsvm`_ and `liblinear`_ use ``C`` as
regularization parameter, most other estimators use ``alpha``. The exact
equivalence between the amount of regularization of two models depends on
the exact objective function optimized by the model. For example, when the
estimator used is :class:`~sklearn.linear_model.Ridge` regression,
the relation between them is given as :math:`C = \frac{1}{alpha}`.
|details-start|
**LinearSVC**
|details-split|
The primal problem can be equivalently formulated as
.. math::
\min_ {w, b} \frac{1}{2} w^T w + C \sum_{i=1}^{n}\max(0, 1 - y_i (w^T \phi(x_i) + b)),
where we make use of the `hinge loss
`_. This is the form that is
directly optimized by :class:`LinearSVC`, but unlike the dual form, this one
does not involve inner products between samples, so the famous kernel trick
cannot be applied. This is why only the linear kernel is supported by
:class:`LinearSVC` (:math:`\phi` is the identity function).
|details-end|
.. _nu_svc:
|details-start|
**NuSVC**
|details-split|
The :math:`\nu`-SVC formulation [#7]_ is a reparameterization of the
:math:`C`-SVC and therefore mathematically equivalent.
We introduce a new parameter :math:`\nu` (instead of :math:`C`) which
controls the number of support vectors and *margin errors*:
:math:`\nu \in (0, 1]` is an upper bound on the fraction of margin errors and
a lower bound of the fraction of support vectors. A margin error corresponds
to a sample that lies on the wrong side of its margin boundary: it is either
misclassified, or it is correctly classified but does not lie beyond the
margin.
|details-end|
SVR
---
Given training vectors :math:`x_i \in \mathbb{R}^p`, i=1,..., n, and a
vector :math:`y \in \mathbb{R}^n` :math:`\varepsilon`-SVR solves the following primal problem:
.. math::
\min_ {w, b, \zeta, \zeta^*} \frac{1}{2} w^T w + C \sum_{i=1}^{n} (\zeta_i + \zeta_i^*)
\textrm {subject to } & y_i - w^T \phi (x_i) - b \leq \varepsilon + \zeta_i,\\
& w^T \phi (x_i) + b - y_i \leq \varepsilon + \zeta_i^*,\\
& \zeta_i, \zeta_i^* \geq 0, i=1, ..., n
Here, we are penalizing samples whose prediction is at least :math:`\varepsilon`
away from their true target. These samples penalize the objective by
:math:`\zeta_i` or :math:`\zeta_i^*`, depending on whether their predictions
lie above or below the :math:`\varepsilon` tube.
The dual problem is
.. math::
\min_{\alpha, \alpha^*} \frac{1}{2} (\alpha - \alpha^*)^T Q (\alpha - \alpha^*) + \varepsilon e^T (\alpha + \alpha^*) - y^T (\alpha - \alpha^*)
\textrm {subject to } & e^T (\alpha - \alpha^*) = 0\\
& 0 \leq \alpha_i, \alpha_i^* \leq C, i=1, ..., n
where :math:`e` is the vector of all ones,
:math:`Q` is an :math:`n` by :math:`n` positive semidefinite matrix,
:math:`Q_{ij} \equiv K(x_i, x_j) = \phi (x_i)^T \phi (x_j)`
is the kernel. Here training vectors are implicitly mapped into a higher
(maybe infinite) dimensional space by the function :math:`\phi`.
The prediction is:
.. math:: \sum_{i \in SV}(\alpha_i - \alpha_i^*) K(x_i, x) + b
These parameters can be accessed through the attributes ``dual_coef_``
which holds the difference :math:`\alpha_i - \alpha_i^*`, ``support_vectors_`` which
holds the support vectors, and ``intercept_`` which holds the independent
term :math:`b`
|details-start|
**LinearSVR**
|details-split|
The primal problem can be equivalently formulated as
.. math::
\min_ {w, b} \frac{1}{2} w^T w + C \sum_{i=1}^{n}\max(0, |y_i - (w^T \phi(x_i) + b)| - \varepsilon),
where we make use of the epsilon-insensitive loss, i.e. errors of less than
:math:`\varepsilon` are ignored. This is the form that is directly optimized
by :class:`LinearSVR`.
|details-end|
.. _svm_implementation_details:
Implementation details
======================
Internally, we use `libsvm`_ [#4]_ and `liblinear`_ [#3]_ to handle all
computations. These libraries are wrapped using C and Cython.
For a description of the implementation and details of the algorithms
used, please refer to their respective papers.
.. _`libsvm`: https://www.csie.ntu.edu.tw/~cjlin/libsvm/
.. _`liblinear`: https://www.csie.ntu.edu.tw/~cjlin/liblinear/
.. topic:: References:
.. [#1] Platt `"Probabilistic outputs for SVMs and comparisons to
regularized likelihood methods"
`_.
.. [#2] Wu, Lin and Weng, `"Probability estimates for multi-class
classification by pairwise coupling"
`_, JMLR
5:975-1005, 2004.
.. [#3] Fan, Rong-En, et al.,
`"LIBLINEAR: A library for large linear classification."
`_,
Journal of machine learning research 9.Aug (2008): 1871-1874.
.. [#4] Chang and Lin, `LIBSVM: A Library for Support Vector Machines
`_.
.. [#5] Bishop, `Pattern recognition and machine learning
`_,
chapter 7 Sparse Kernel Machines
.. [#6] :doi:`"A Tutorial on Support Vector Regression"
<10.1023/B:STCO.0000035301.49549.88>`
Alex J. Smola, Bernhard Schölkopf - Statistics and Computing archive
Volume 14 Issue 3, August 2004, p. 199-222.
.. [#7] Schölkopf et. al `New Support Vector Algorithms
`_
.. [#8] Crammer and Singer `On the Algorithmic Implementation ofMulticlass
Kernel-based Vector Machines
`_,
JMLR 2001.