PLearn 0.1
|
00001 // -*- C++ -*- 00002 00003 // SoftmaxModule.cc 00004 // 00005 // Copyright (C) 2006 Pascal Lamblin 00006 // 00007 // Redistribution and use in source and binary forms, with or without 00008 // modification, are permitted provided that the following conditions are met: 00009 // 00010 // 1. Redistributions of source code must retain the above copyright 00011 // notice, this list of conditions and the following disclaimer. 00012 // 00013 // 2. Redistributions in binary form must reproduce the above copyright 00014 // notice, this list of conditions and the following disclaimer in the 00015 // documentation and/or other materials provided with the distribution. 00016 // 00017 // 3. The name of the authors may not be used to endorse or promote 00018 // products derived from this software without specific prior written 00019 // permission. 00020 // 00021 // THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR 00022 // IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 00023 // OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN 00024 // NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 00025 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED 00026 // TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 00027 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 00028 // LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 00029 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 00030 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 00031 // 00032 // This file is part of the PLearn library. For more information on the PLearn 00033 // library, go to the PLearn Web site at www.plearn.org 00034 00035 // Authors: Pascal Lamblin 00036 00041 #include "SoftmaxModule.h" 00042 #include <plearn/math/TMat_maths.h> 00043 00044 namespace PLearn { 00045 using namespace std; 00046 00047 PLEARN_IMPLEMENT_OBJECT( 00048 SoftmaxModule, 00049 "Computes the softmax function on a vector.", 00050 "" 00051 ); 00052 00054 // SoftmaxModule // 00056 SoftmaxModule::SoftmaxModule() 00057 {} 00058 00060 // declareOptions // 00062 void SoftmaxModule::declareOptions(OptionList& ol) 00063 { 00064 // Now call the parent class' declareOptions 00065 inherited::declareOptions(ol); 00066 00067 // Hide unused options. 00068 00069 redeclareOption(ol, "output_size", &SoftmaxModule::output_size, 00070 OptionBase::nosave, 00071 "Set at build time."); 00072 } 00073 00075 // build_ // 00077 void SoftmaxModule::build_() 00078 { 00079 output_size = input_size; 00080 } 00081 00083 // build // 00085 void SoftmaxModule::build() 00086 { 00087 inherited::build(); 00088 build_(); 00089 } 00090 00092 // makeDeepCopyFromShallowCopy // 00094 void SoftmaxModule::makeDeepCopyFromShallowCopy(CopiesMap& copies) 00095 { 00096 inherited::makeDeepCopyFromShallowCopy(copies); 00097 } 00098 00100 // fprop // 00102 void SoftmaxModule::fprop(const Vec& input, Vec& output) const 00103 { 00104 PLASSERT( input.size() == input_size ); 00105 output.resize( output_size ); 00106 00107 softmax( input, output ); 00108 } 00109 00110 void SoftmaxModule::fprop(const Mat& inputs, Mat& outputs) 00111 { 00112 PLASSERT( inputs.width() == input_size ); 00113 int n = inputs.length(); 00114 outputs.resize(n, output_size ); 00115 for (int i = 0; i < n; i++) 00116 softmax(inputs(i), outputs(i)); 00117 } 00118 00120 // bpropUpdate // 00122 void SoftmaxModule::bpropUpdate(const Vec& input, const Vec& output, 00123 Vec& input_gradient, 00124 const Vec& output_gradient, 00125 bool accumulate) 00126 { 00127 PLASSERT( input.size() == input_size ); 00128 PLASSERT( output.size() == output_size ); 00129 PLASSERT( output_gradient.size() == output_size ); 00130 00131 if( accumulate ) 00132 { 00133 PLASSERT_MSG( input_gradient.size() == input_size, 00134 "Cannot resize input_gradient AND accumulate into it" ); 00135 } 00136 else 00137 { 00138 input_gradient.resize( input_size ); 00139 input_gradient.clear(); 00140 } 00141 00142 // input_gradient[i] = output_gradient[i] * output[i] 00143 // - (output_gradient . output ) output[i] 00144 real outg_dot_out = dot( output_gradient, output ); 00145 for( int i=0 ; i<input_size ; i++ ) 00146 { 00147 real in_grad_i = (output_gradient[i] - outg_dot_out) * output[i]; 00148 input_gradient[i] += in_grad_i; 00149 } 00150 } 00151 00152 void SoftmaxModule::bpropUpdate(const Mat& inputs, const Mat& outputs, 00153 Mat& input_gradients, 00154 const Mat& output_gradients, 00155 bool accumulate) 00156 { 00157 PLASSERT( inputs.width() == input_size ); 00158 PLASSERT( outputs.width() == output_size ); 00159 PLASSERT( output_gradients.width() == output_size ); 00160 00161 if( accumulate ) 00162 { 00163 PLASSERT_MSG( input_gradients.width() == input_size && 00164 input_gradients.length() == inputs.length(), 00165 "Cannot resize input_gradients and accumulate into it" ); 00166 } 00167 else 00168 { 00169 input_gradients.resize(inputs.length(), input_size); 00170 input_gradients.fill(0); 00171 } 00172 00173 for (int j = 0; j < inputs.length(); j++) { 00174 // input_gradient[i] = output_gradient[i] * output[i] 00175 // - (output_gradient . output ) output[i] 00176 real outg_dot_out = dot(output_gradients(j), outputs(j)); 00177 for( int i=0 ; i<input_size ; i++ ) 00178 input_gradients(j, i) += 00179 (output_gradients(j, i) - outg_dot_out) * outputs(j, i); 00180 } 00181 } 00182 00184 // forget // 00186 void SoftmaxModule::forget() 00187 { 00188 } 00189 00191 // setLearningRate // 00193 void SoftmaxModule::setLearningRate(real dynamic_learning_rate) 00194 { 00195 } 00196 00198 // bbpropUpdate // 00200 void SoftmaxModule::bbpropUpdate(const Vec& input, const Vec& output, 00201 Vec& input_gradient, 00202 const Vec& output_gradient, 00203 Vec& input_diag_hessian, 00204 const Vec& output_diag_hessian, 00205 bool accumulate) 00206 { 00207 PLERROR( "Not implemented yet, please come back later or complain to" 00208 " lamblinp." ); 00209 } 00210 00211 00212 } // end of namespace PLearn 00213 00214 00215 /* 00216 Local Variables: 00217 mode:c++ 00218 c-basic-offset:4 00219 c-file-style:"stroustrup" 00220 c-file-offsets:((innamespace . 0)(inline-open . 0)) 00221 indent-tabs-mode:nil 00222 fill-column:79 00223 End: 00224 */ 00225 // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=8:softtabstop=4:encoding=utf-8:textwidth=79 :