3709 |
08 Nov 17 |
peter |
1 |
#ifndef theplu_yat_classifier_perceptron |
3709 |
08 Nov 17 |
peter |
2 |
#define theplu_yat_classifier_perceptron |
3709 |
08 Nov 17 |
peter |
3 |
|
3709 |
08 Nov 17 |
peter |
// $Id$ |
3709 |
08 Nov 17 |
peter |
5 |
|
3709 |
08 Nov 17 |
peter |
6 |
/* |
4207 |
26 Aug 22 |
peter |
Copyright (C) 2017, 2021, 2022 Peter Johansson |
3709 |
08 Nov 17 |
peter |
8 |
|
3709 |
08 Nov 17 |
peter |
This file is part of the yat library, http://dev.thep.lu.se/yat |
3709 |
08 Nov 17 |
peter |
10 |
|
3709 |
08 Nov 17 |
peter |
The yat library is free software; you can redistribute it and/or |
3709 |
08 Nov 17 |
peter |
modify it under the terms of the GNU General Public License as |
3709 |
08 Nov 17 |
peter |
published by the Free Software Foundation; either version 3 of the |
3709 |
08 Nov 17 |
peter |
License, or (at your option) any later version. |
3709 |
08 Nov 17 |
peter |
15 |
|
3709 |
08 Nov 17 |
peter |
The yat library is distributed in the hope that it will be useful, |
3709 |
08 Nov 17 |
peter |
but WITHOUT ANY WARRANTY; without even the implied warranty of |
3709 |
08 Nov 17 |
peter |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
3709 |
08 Nov 17 |
peter |
General Public License for more details. |
3709 |
08 Nov 17 |
peter |
20 |
|
3709 |
08 Nov 17 |
peter |
You should have received a copy of the GNU General Public License |
3709 |
08 Nov 17 |
peter |
along with yat. If not, see <http://www.gnu.org/licenses/>. |
3709 |
08 Nov 17 |
peter |
23 |
*/ |
3709 |
08 Nov 17 |
peter |
24 |
|
3709 |
08 Nov 17 |
peter |
25 |
#include <yat/utility/Matrix.h> |
3709 |
08 Nov 17 |
peter |
26 |
#include <yat/utility/Vector.h> |
3709 |
08 Nov 17 |
peter |
27 |
|
3709 |
08 Nov 17 |
peter |
28 |
namespace theplu { |
3709 |
08 Nov 17 |
peter |
29 |
namespace yat { |
3709 |
08 Nov 17 |
peter |
30 |
namespace classifier { |
3709 |
08 Nov 17 |
peter |
31 |
|
3709 |
08 Nov 17 |
peter |
32 |
class Target; |
3709 |
08 Nov 17 |
peter |
33 |
|
3709 |
08 Nov 17 |
peter |
34 |
/** |
3709 |
08 Nov 17 |
peter |
\brief A Single-layer Perceptron |
3709 |
08 Nov 17 |
peter |
36 |
|
4052 |
26 Mar 21 |
peter |
Data are modeled as \f$ \mu = \frac{1}{1 + \exp(-wx)} \f$ |
3709 |
08 Nov 17 |
peter |
38 |
|
3709 |
08 Nov 17 |
peter |
\since New in yat 0.16 |
3709 |
08 Nov 17 |
peter |
40 |
*/ |
3709 |
08 Nov 17 |
peter |
41 |
class Perceptron |
3709 |
08 Nov 17 |
peter |
42 |
{ |
3709 |
08 Nov 17 |
peter |
43 |
public: |
3709 |
08 Nov 17 |
peter |
44 |
/** |
3709 |
08 Nov 17 |
peter |
Estimated covariance of weight vector. |
3709 |
08 Nov 17 |
peter |
46 |
|
3709 |
08 Nov 17 |
peter |
Covariance is estimated as \f$ \left(X'SX\right)^{-1} \f$ where |
3709 |
08 Nov 17 |
peter |
\f$ S \f$ is a diagnoal matrix with \f$ S_{ii} = \mu_i |
3709 |
08 Nov 17 |
peter |
(1-\mu_i) \f$ where \f$ \mu_i \f$ is the expected value of |
3709 |
08 Nov 17 |
peter |
sample i. |
3709 |
08 Nov 17 |
peter |
51 |
*/ |
3709 |
08 Nov 17 |
peter |
52 |
const utility::Matrix& covariance(void) const; |
3709 |
08 Nov 17 |
peter |
53 |
|
3709 |
08 Nov 17 |
peter |
54 |
/** |
3709 |
08 Nov 17 |
peter |
The odds ratio is defined as \f$ \textrm{OR} = \exp(w_i) \f$ |
3709 |
08 Nov 17 |
peter |
56 |
*/ |
3709 |
08 Nov 17 |
peter |
57 |
double oddsratio(size_t i) const; |
3709 |
08 Nov 17 |
peter |
58 |
|
3709 |
08 Nov 17 |
peter |
59 |
/** |
3709 |
08 Nov 17 |
peter |
The lower end of the confidence interval of estimation of |
3709 |
08 Nov 17 |
peter |
oddsratio \a i with confidence 1 - \a alpha. The true value is |
3709 |
08 Nov 17 |
peter |
estimated to be within confidence interval with probability 1 - |
3709 |
08 Nov 17 |
peter |
\a alpha. |
3709 |
08 Nov 17 |
peter |
64 |
*/ |
3709 |
08 Nov 17 |
peter |
65 |
double oddsratio_lower_CI(size_t i, double alpha=0.05) const; |
3709 |
08 Nov 17 |
peter |
66 |
|
3709 |
08 Nov 17 |
peter |
67 |
/** |
3709 |
08 Nov 17 |
peter |
The lower end of the confidence interval of estimation of |
3709 |
08 Nov 17 |
peter |
oddsratio \a i with confidence 1 - \a alpha. The true value is |
3709 |
08 Nov 17 |
peter |
estimated to be within confidence interval with probability 1 - |
3709 |
08 Nov 17 |
peter |
\a alpha. |
3709 |
08 Nov 17 |
peter |
72 |
*/ |
3709 |
08 Nov 17 |
peter |
73 |
double oddsratio_upper_CI(size_t i, double alpha=0.05) const; |
3709 |
08 Nov 17 |
peter |
74 |
|
3709 |
08 Nov 17 |
peter |
75 |
/** |
3709 |
08 Nov 17 |
peter |
\return p-value that for null hypothesis that ith weight is zero |
3709 |
08 Nov 17 |
peter |
77 |
*/ |
3709 |
08 Nov 17 |
peter |
78 |
double p_value(size_t i) const; |
3709 |
08 Nov 17 |
peter |
79 |
|
3709 |
08 Nov 17 |
peter |
80 |
/** |
3709 |
08 Nov 17 |
peter |
\return \f$ \frac{1}{1 + \exp(-wx)} \f$ |
3709 |
08 Nov 17 |
peter |
82 |
*/ |
3709 |
08 Nov 17 |
peter |
83 |
double predict(const utility::VectorBase& x) const; |
3709 |
08 Nov 17 |
peter |
84 |
|
3709 |
08 Nov 17 |
peter |
85 |
/** |
3709 |
08 Nov 17 |
peter |
\brief train the model |
3709 |
08 Nov 17 |
peter |
87 |
|
4052 |
26 Mar 21 |
peter |
Model parameters, \f$w\f$, are calculated such that the |
4200 |
19 Aug 22 |
peter |
log-likelihood, |
4052 |
26 Mar 21 |
peter |
\f$ \log \mathcal{L} = \sum y_i \log \left(\mu_i\right) + |
4052 |
26 Mar 21 |
peter |
(1-y_i) \log \left(1 - \mu_i\right) \f$, |
4052 |
26 Mar 21 |
peter |
is maximized. |
4052 |
26 Mar 21 |
peter |
93 |
|
3709 |
08 Nov 17 |
peter |
\param x each row corresponds to a data point and each column a |
3709 |
08 Nov 17 |
peter |
feature. |
3709 |
08 Nov 17 |
peter |
\param target describes the class label for each data |
3709 |
08 Nov 17 |
peter |
point. Data that has binary set are trained to output 1. |
3709 |
08 Nov 17 |
peter |
98 |
*/ |
4125 |
14 Jan 22 |
peter |
99 |
void train(const utility::MatrixBase& x, const Target& target); |
3709 |
08 Nov 17 |
peter |
100 |
|
3709 |
08 Nov 17 |
peter |
101 |
/** |
3709 |
08 Nov 17 |
peter |
\return trained weight vector, \f$w\f$. |
3709 |
08 Nov 17 |
peter |
103 |
*/ |
3709 |
08 Nov 17 |
peter |
104 |
const utility::Vector& weight(void) const; |
3709 |
08 Nov 17 |
peter |
105 |
private: |
3709 |
08 Nov 17 |
peter |
106 |
utility::Vector weight_; |
3709 |
08 Nov 17 |
peter |
107 |
utility::Matrix covariance_; |
3709 |
08 Nov 17 |
peter |
108 |
double margin(size_t i, double alpha) const; |
3709 |
08 Nov 17 |
peter |
// using compiler generated copy |
3709 |
08 Nov 17 |
peter |
//Perceptron(const Perceptron&) |
3709 |
08 Nov 17 |
peter |
//Perceptron& operator=(const Perceptron&) |
3709 |
08 Nov 17 |
peter |
112 |
}; |
3709 |
08 Nov 17 |
peter |
113 |
|
3709 |
08 Nov 17 |
peter |
114 |
|
3709 |
08 Nov 17 |
peter |
115 |
}}}// end of namespace classifier, yat, and theplu |
3709 |
08 Nov 17 |
peter |
116 |
#endif |