| 1 |
/* |
| 2 |
* Software License Agreement (BSD License) |
| 3 |
* |
| 4 |
* Copyright (c) 2013-2014, Willow Garage, Inc. |
| 5 |
* Copyright (c) 2014-2016, Open Source Robotics Foundation |
| 6 |
* All rights reserved. |
| 7 |
* |
| 8 |
* Redistribution and use in source and binary forms, with or without |
| 9 |
* modification, are permitted provided that the following conditions |
| 10 |
* are met: |
| 11 |
* |
| 12 |
* * Redistributions of source code must retain the above copyright |
| 13 |
* notice, this list of conditions and the following disclaimer. |
| 14 |
* * Redistributions in binary form must reproduce the above |
| 15 |
* copyright notice, this list of conditions and the following |
| 16 |
* disclaimer in the documentation and/or other materials provided |
| 17 |
* with the distribution. |
| 18 |
* * Neither the name of Open Source Robotics Foundation nor the names of its |
| 19 |
* contributors may be used to endorse or promote products derived |
| 20 |
* from this software without specific prior written permission. |
| 21 |
* |
| 22 |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 23 |
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 24 |
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
| 25 |
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
| 26 |
* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 27 |
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
| 28 |
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 29 |
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
| 30 |
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 31 |
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN |
| 32 |
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
| 33 |
* POSSIBILITY OF SUCH DAMAGE. |
| 34 |
*/ |
| 35 |
|
| 36 |
/** \author Jia Pan */ |
| 37 |
|
| 38 |
#ifndef FCL_LEARNING_CLASSIFIER_H |
| 39 |
#define FCL_LEARNING_CLASSIFIER_H |
| 40 |
|
| 41 |
#include <vector> |
| 42 |
|
| 43 |
#include "fcl/common/data_types.h" |
| 44 |
|
| 45 |
namespace fcl |
| 46 |
{ |
| 47 |
|
| 48 |
template<typename S, std::size_t N> |
| 49 |
struct Item |
| 50 |
{ |
| 51 |
VectorN<S, N> q; |
| 52 |
bool label; |
| 53 |
S w; |
| 54 |
|
| 55 |
Item(const VectorN<S, N>& q_, bool label_, S w_ = 1); |
| 56 |
|
| 57 |
Item(); |
| 58 |
|
| 59 |
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(S, N) |
| 60 |
}; |
| 61 |
|
| 62 |
template<typename S, std::size_t N> |
| 63 |
struct Scaler |
| 64 |
{ |
| 65 |
VectorN<S, N> v_min, v_max; |
| 66 |
Scaler(); |
| 67 |
|
| 68 |
Scaler(const VectorN<S, N>& v_min_, const VectorN<S, N>& v_max_); |
| 69 |
|
| 70 |
VectorN<S, N> scale(const VectorN<S, N>& v) const; |
| 71 |
|
| 72 |
VectorN<S, N> unscale(const VectorN<S, N>& v) const; |
| 73 |
|
| 74 |
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(S, N) |
| 75 |
}; |
| 76 |
|
| 77 |
template<typename S> |
| 78 |
struct PredictResult |
| 79 |
{ |
| 80 |
bool label; |
| 81 |
S prob; |
| 82 |
|
| 83 |
PredictResult(); |
| 84 |
PredictResult(bool label_, S prob_ = 1); |
| 85 |
}; |
| 86 |
|
| 87 |
template<typename S, std::size_t N> |
| 88 |
class SVMClassifier |
| 89 |
{ |
| 90 |
public: |
| 91 |
|
| 92 |
~SVMClassifier(); |
| 93 |
|
| 94 |
virtual PredictResult<S> predict(const VectorN<S, N>& q) const = 0; |
| 95 |
virtual std::vector<PredictResult<S>> predict( |
| 96 |
const std::vector<VectorN<S, N> >& qs) const = 0; |
| 97 |
|
| 98 |
virtual std::vector<Item<S, N> > getSupportVectors() const = 0; |
| 99 |
virtual void setScaler(const Scaler<S, N>& scaler) = 0; |
| 100 |
|
| 101 |
virtual void learn(const std::vector<Item<S, N> >& data) = 0; |
| 102 |
|
| 103 |
S error_rate(const std::vector<Item<S, N> >& data) const; |
| 104 |
}; |
| 105 |
|
| 106 |
template<typename S, std::size_t N> |
| 107 |
Scaler<S, N> computeScaler(const std::vector<Item<S, N> >& data); |
| 108 |
|
| 109 |
template<typename S, std::size_t N> |
| 110 |
Scaler<S, N> computeScaler(const std::vector<VectorN<S, N> >& data); |
| 111 |
|
| 112 |
//============================================================================// |
| 113 |
// // |
| 114 |
// Implementations // |
| 115 |
// // |
| 116 |
//============================================================================// |
| 117 |
|
| 118 |
//============================================================================== |
| 119 |
template<typename S, std::size_t N> |
| 120 |
Item<S, N>::Item(const VectorN<S, N>& q_, bool label_, S w_) |
| 121 |
: q(q_), label(label_), w(w_) |
| 122 |
{ |
| 123 |
// Do nothing |
| 124 |
} |
| 125 |
|
| 126 |
//============================================================================== |
| 127 |
template<typename S, std::size_t N> |
| 128 |
Item<S, N>::Item() |
| 129 |
{ |
| 130 |
// Do nothing |
| 131 |
} |
| 132 |
|
| 133 |
//============================================================================== |
| 134 |
template<typename S, std::size_t N> |
| 135 |
Scaler<S, N>::Scaler() |
| 136 |
{ |
| 137 |
// default no scale |
| 138 |
for(std::size_t i = 0; i < N; ++i) |
| 139 |
{ |
| 140 |
v_min[i] = 0; |
| 141 |
v_max[i] = 1; |
| 142 |
} |
| 143 |
} |
| 144 |
|
| 145 |
//============================================================================== |
| 146 |
template<typename S, std::size_t N> |
| 147 |
Scaler<S, N>::Scaler(const VectorN<S, N>& v_min_, const VectorN<S, N>& v_max_) |
| 148 |
: v_min(v_min_), v_max(v_max_) |
| 149 |
{ |
| 150 |
// Do nothing |
| 151 |
} |
| 152 |
|
| 153 |
//============================================================================== |
| 154 |
template<typename S, std::size_t N> |
| 155 |
VectorN<S, N> Scaler<S, N>::scale(const VectorN<S, N>& v) const |
| 156 |
{ |
| 157 |
VectorN<S, N> res; |
| 158 |
for(std::size_t i = 0; i < N; ++i) |
| 159 |
res[i] = (v[i] - v_min[i]) / (v_max[i] - v_min[i]); |
| 160 |
return res; |
| 161 |
} |
| 162 |
|
| 163 |
//============================================================================== |
| 164 |
template<typename S, std::size_t N> |
| 165 |
VectorN<S, N> Scaler<S, N>::unscale(const VectorN<S, N>& v) const |
| 166 |
{ |
| 167 |
VectorN<S, N> res; |
| 168 |
for(std::size_t i = 0; i < N; ++i) |
| 169 |
res[i] = v[i] * (v_max[i] - v_min[i]) + v_min[i]; |
| 170 |
return res; |
| 171 |
} |
| 172 |
|
| 173 |
//============================================================================== |
| 174 |
template<typename S> |
| 175 |
PredictResult<S>::PredictResult() |
| 176 |
{ |
| 177 |
// Do nothing |
| 178 |
} |
| 179 |
|
| 180 |
//============================================================================== |
| 181 |
template<typename S> |
| 182 |
PredictResult<S>::PredictResult(bool label_, S prob_) |
| 183 |
: label(label_), prob(prob_) |
| 184 |
{ |
| 185 |
// Do nothing |
| 186 |
} |
| 187 |
|
| 188 |
//============================================================================== |
| 189 |
template<typename S, std::size_t N> |
| 190 |
SVMClassifier<S, N>::~SVMClassifier() |
| 191 |
{ |
| 192 |
// Do nothing |
| 193 |
} |
| 194 |
|
| 195 |
//============================================================================== |
| 196 |
template<typename S, std::size_t N> |
| 197 |
S SVMClassifier<S, N>::error_rate(const std::vector<Item<S, N> >& data) const |
| 198 |
{ |
| 199 |
std::size_t num = data.size(); |
| 200 |
|
| 201 |
std::size_t error_num = 0; |
| 202 |
for(std::size_t i = 0; i < data.size(); ++i) |
| 203 |
{ |
| 204 |
PredictResult<S> res = predict(data[i].q); |
| 205 |
if(res.label != data[i].label) |
| 206 |
error_num++; |
| 207 |
} |
| 208 |
|
| 209 |
return error_num / (S)num; |
| 210 |
} |
| 211 |
|
| 212 |
//============================================================================== |
| 213 |
template<typename S, std::size_t N> |
| 214 |
Scaler<S, N> computeScaler(const std::vector<Item<S, N> >& data) |
| 215 |
{ |
| 216 |
VectorN<S, N> lower_bound, upper_bound; |
| 217 |
for(std::size_t j = 0; j < N; ++j) |
| 218 |
{ |
| 219 |
lower_bound[j] = std::numeric_limits<S>::max(); |
| 220 |
upper_bound[j] = -std::numeric_limits<S>::max(); |
| 221 |
} |
| 222 |
|
| 223 |
for(std::size_t i = 0; i < data.size(); ++i) |
| 224 |
{ |
| 225 |
for(std::size_t j = 0; j < N; ++j) |
| 226 |
{ |
| 227 |
if(data[i].q[j] < lower_bound[j]) lower_bound[j] = data[i].q[j]; |
| 228 |
if(data[i].q[j] > upper_bound[j]) upper_bound[j] = data[i].q[j]; |
| 229 |
} |
| 230 |
} |
| 231 |
|
| 232 |
return Scaler<S, N>(lower_bound, upper_bound); |
| 233 |
} |
| 234 |
|
| 235 |
//============================================================================== |
| 236 |
template<typename S, std::size_t N> |
| 237 |
Scaler<S, N> computeScaler(const std::vector<VectorN<S, N> >& data) |
| 238 |
{ |
| 239 |
VectorN<S, N> lower_bound, upper_bound; |
| 240 |
for(std::size_t j = 0; j < N; ++j) |
| 241 |
{ |
| 242 |
lower_bound[j] = std::numeric_limits<S>::max(); |
| 243 |
upper_bound[j] = -std::numeric_limits<S>::max(); |
| 244 |
} |
| 245 |
|
| 246 |
for(std::size_t i = 0; i < data.size(); ++i) |
| 247 |
{ |
| 248 |
for(std::size_t j = 0; j < N; ++j) |
| 249 |
{ |
| 250 |
if(data[i][j] < lower_bound[j]) lower_bound[j] = data[i][j]; |
| 251 |
if(data[i][j] > upper_bound[j]) upper_bound[j] = data[i][j]; |
| 252 |
} |
| 253 |
} |
| 254 |
|
| 255 |
return Scaler<S, N>(lower_bound, upper_bound); |
| 256 |
} |
| 257 |
|
| 258 |
} // namespace fcl |
| 259 |
|
| 260 |
#endif |