<html><!-- Created using the cpp_pretty_printer from the dlib C++ library. See http://dlib.net for updates. --><head><title>dlib C++ Library - rr_trainer_abstract.h</title></head><body bgcolor='white'><pre>
<font color='#009900'>// Copyright (C) 2010 Davis E. King (davis@dlib.net)
</font><font color='#009900'>// License: Boost Software License See LICENSE.txt for the full license.
</font><font color='#0000FF'>#undef</font> DLIB_RR_TRAInER_ABSTRACT_Hh_
<font color='#0000FF'>#ifdef</font> DLIB_RR_TRAInER_ABSTRACT_Hh_
<font color='#0000FF'>#include</font> "<a style='text-decoration:none' href='../algs.h.html'>../algs.h</a>"
<font color='#0000FF'>#include</font> "<a style='text-decoration:none' href='function_abstract.h.html'>function_abstract.h</a>"
<font color='#0000FF'>namespace</font> dlib
<b>{</b>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> K
<font color='#5555FF'>></font>
<font color='#0000FF'>class</font> <b><a name='rr_trainer'></a>rr_trainer</b>
<b>{</b>
<font color='#009900'>/*!
REQUIREMENTS ON K
is the dlib::linear_kernel instantiated with some kind of column vector.
INITIAL VALUE
- get_lambda() == 0
- will_use_regression_loss_for_loo_cv() == true
- get_search_lambdas() == logspace(-9, 2, 50)
- this object will not be verbose unless be_verbose() is called
WHAT THIS OBJECT REPRESENTS
This object represents a tool for performing linear ridge regression
(This basic algorithm is also known my many other names, e.g. regularized
least squares or least squares SVM).
The exact definition of what this algorithm does is this:
Find w and b that minimizes the following (x_i are input samples and y_i are target values):
lambda*dot(w,w) + sum_over_i( (f(x_i) - y_i)^2 )
where f(x) == dot(x,w) - b
So this algorithm is just regular old least squares regression but
with the addition of a regularization term which encourages small w.
It is capable of estimating the lambda parameter using leave-one-out cross-validation.
The leave-one-out cross-validation implementation is based on the techniques
discussed in this paper:
Notes on Regularized Least Squares by Ryan M. Rifkin and Ross A. Lippert.
!*/</font>
<font color='#0000FF'>public</font>:
<font color='#0000FF'>typedef</font> K kernel_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'>typename</font> kernel_type::scalar_type scalar_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'>typename</font> kernel_type::sample_type sample_type;
<font color='#0000FF'>typedef</font> <font color='#0000FF'>typename</font> kernel_type::mem_manager_type mem_manager_type;
<font color='#0000FF'>typedef</font> decision_function<font color='#5555FF'><</font>kernel_type<font color='#5555FF'>></font> trained_function_type;
<b><a name='rr_trainer'></a>rr_trainer</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- This object is properly initialized and ready to be used.
!*/</font>
<font color='#0000FF'><u>void</u></font> <b><a name='be_verbose'></a>be_verbose</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- This object will print status messages to standard out.
!*/</font>
<font color='#0000FF'><u>void</u></font> <b><a name='be_quiet'></a>be_quiet</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- this object will not print anything to standard out
!*/</font>
<font color='#0000FF'>const</font> kernel_type <b><a name='get_kernel'></a>get_kernel</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns a copy of the kernel function in use by this object. Since
the linear kernels don't have any parameters this function just
returns kernel_type()
!*/</font>
<font color='#0000FF'><u>void</u></font> <b><a name='set_lambda'></a>set_lambda</b> <font face='Lucida Console'>(</font>
scalar_type lambda
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- lambda >= 0
ensures
- #get_lambda() == lambda
!*/</font>
<font color='#0000FF'>const</font> scalar_type <b><a name='get_lambda'></a>get_lambda</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns the regularization parameter. It is the parameter that
determines the trade off between trying to fit the training data
exactly or allowing more errors but hopefully improving the
generalization ability of the resulting function. Smaller values
encourage exact fitting while larger values of lambda may encourage
better generalization.
Note that a lambda of 0 has a special meaning. It indicates to this
object that it should automatically determine an appropriate lambda
value. This is done using leave-one-out cross-validation.
!*/</font>
<font color='#0000FF'><u>void</u></font> <b><a name='use_regression_loss_for_loo_cv'></a>use_regression_loss_for_loo_cv</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- #will_use_regression_loss_for_loo_cv() == true
!*/</font>
<font color='#0000FF'><u>void</u></font> <b><a name='use_classification_loss_for_loo_cv'></a>use_classification_loss_for_loo_cv</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
ensures
- #will_use_regression_loss_for_loo_cv() == false
!*/</font>
<font color='#0000FF'><u>bool</u></font> <b><a name='will_use_regression_loss_for_loo_cv'></a>will_use_regression_loss_for_loo_cv</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns true if the automatic lambda estimation will attempt to estimate a lambda
appropriate for a regression task. Otherwise it will try and find one which
minimizes the number of classification errors.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> EXP<font color='#5555FF'>></font>
<font color='#0000FF'><u>void</u></font> <b><a name='set_search_lambdas'></a>set_search_lambdas</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> matrix_exp<font color='#5555FF'><</font>EXP<font color='#5555FF'>></font><font color='#5555FF'>&</font> lambdas
<font face='Lucida Console'>)</font>;
<font color='#009900'>/*!
requires
- is_vector(lambdas) == true
- lambdas.size() > 0
- min(lambdas) > 0
- lambdas must contain floating point numbers
ensures
- #get_search_lambdas() == lambdas
!*/</font>
<font color='#0000FF'>const</font> matrix<font color='#5555FF'><</font>scalar_type,<font color='#979000'>0</font>,<font color='#979000'>0</font>,mem_manager_type<font color='#5555FF'>></font><font color='#5555FF'>&</font> <b><a name='get_search_lambdas'></a>get_search_lambdas</b> <font face='Lucida Console'>(</font>
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
ensures
- returns a matrix M such that:
- is_vector(M) == true
- M == a list of all the lambda values which will be tried when performing
LOO cross-validation for determining the best lambda.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> in_sample_vector_type,
<font color='#0000FF'>typename</font> in_scalar_vector_type
<font color='#5555FF'>></font>
<font color='#0000FF'>const</font> decision_function<font color='#5555FF'><</font>kernel_type<font color='#5555FF'>></font> <b><a name='train'></a>train</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> in_sample_vector_type<font color='#5555FF'>&</font> x,
<font color='#0000FF'>const</font> in_scalar_vector_type<font color='#5555FF'>&</font> y
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
requires
- x == a matrix or something convertible to a matrix via mat().
Also, x should contain sample_type objects.
- y == a matrix or something convertible to a matrix via mat().
Also, y should contain scalar_type objects.
- is_learning_problem(x,y) == true
- if (get_lambda() == 0 && will_use_regression_loss_for_loo_cv() == false) then
- is_binary_classification_problem(x,y) == true
(i.e. if you want this algorithm to estimate a lambda appropriate for
classification functions then you had better give a valid classification
problem)
ensures
- performs linear ridge regression given the training samples in x and target values in y.
- returns a decision_function F with the following properties:
- F(new_x) == predicted y value
- F.alpha.size() == 1
- F.basis_vectors.size() == 1
- F.alpha(0) == 1
- if (get_lambda() == 0) then
- This object will perform internal leave-one-out cross-validation to determine an
appropriate lambda automatically. It will compute the LOO error for each lambda
in get_search_lambdas() and select the best one.
- if (will_use_regression_loss_for_loo_cv()) then
- the lambda selected will be the one that minimizes the mean squared error.
- else
- the lambda selected will be the one that minimizes the number classification
mistakes. We say a point is classified correctly if the output of the
decision_function has the same sign as its label.
- #get_lambda() == 0
(i.e. we don't change the get_lambda() value. If you want to know what the
automatically selected lambda value was then call the version of train()
defined below)
- else
- The user supplied value of get_lambda() will be used to perform the ridge regression.
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> in_sample_vector_type,
<font color='#0000FF'>typename</font> in_scalar_vector_type
<font color='#5555FF'>></font>
<font color='#0000FF'>const</font> decision_function<font color='#5555FF'><</font>kernel_type<font color='#5555FF'>></font> <b><a name='train'></a>train</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> in_sample_vector_type<font color='#5555FF'>&</font> x,
<font color='#0000FF'>const</font> in_scalar_vector_type<font color='#5555FF'>&</font> y,
std::vector<font color='#5555FF'><</font>scalar_type<font color='#5555FF'>></font><font color='#5555FF'>&</font> loo_values
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
requires
- all the requirements for train(x,y) must be satisfied
ensures
- returns train(x,y)
(i.e. executes train(x,y) and returns its result)
- #loo_values.size() == y.size()
- for all valid i:
- #loo_values[i] == leave-one-out prediction for the value of y(i) based
on all the training samples other than (x(i),y(i)).
!*/</font>
<font color='#0000FF'>template</font> <font color='#5555FF'><</font>
<font color='#0000FF'>typename</font> in_sample_vector_type,
<font color='#0000FF'>typename</font> in_scalar_vector_type
<font color='#5555FF'>></font>
<font color='#0000FF'>const</font> decision_function<font color='#5555FF'><</font>kernel_type<font color='#5555FF'>></font> <b><a name='train'></a>train</b> <font face='Lucida Console'>(</font>
<font color='#0000FF'>const</font> in_sample_vector_type<font color='#5555FF'>&</font> x,
<font color='#0000FF'>const</font> in_scalar_vector_type<font color='#5555FF'>&</font> y,
std::vector<font color='#5555FF'><</font>scalar_type<font color='#5555FF'>></font><font color='#5555FF'>&</font> loo_values,
scalar_type<font color='#5555FF'>&</font> lambda_used
<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>;
<font color='#009900'>/*!
requires
- all the requirements for train(x,y) must be satisfied
ensures
- returns train(x,y)
(i.e. executes train(x,y) and returns its result)
- #loo_values.size() == y.size()
- for all valid i:
- #loo_values[i] == leave-one-out prediction for the value of y(i) based
on all the training samples other than (x(i),y(i)).
- #lambda_used == the value of lambda used to generate the
decision_function. Note that this lambda value is always
equal to get_lambda() if get_lambda() isn't 0.
!*/</font>
<b>}</b>;
<b>}</b>
<font color='#0000FF'>#endif</font> <font color='#009900'>// DLIB_RR_TRAInER_ABSTRACT_Hh_
</font>
</pre></body></html>