PLearn 0.1
|
00001 // -*- C++ -*- 00002 00003 // PLearn (A C++ Machine Learning Library) 00004 // Copyright (C) 1998 Pascal Vincent 00005 // Copyright (C) 1999-2002 Pascal Vincent, Yoshua Bengio, Rejean Ducharme and University of Montreal 00006 // Copyright (C) 2001-2002 Nicolas Chapados, Ichiro Takeuchi, Jean-Sebastien Senecal 00007 // Copyright (C) 2002 Xiangdong Wang, Christian Dorion 00008 00009 // Redistribution and use in source and binary forms, with or without 00010 // modification, are permitted provided that the following conditions are met: 00011 // 00012 // 1. Redistributions of source code must retain the above copyright 00013 // notice, this list of conditions and the following disclaimer. 00014 // 00015 // 2. Redistributions in binary form must reproduce the above copyright 00016 // notice, this list of conditions and the following disclaimer in the 00017 // documentation and/or other materials provided with the distribution. 00018 // 00019 // 3. The name of the authors may not be used to endorse or promote 00020 // products derived from this software without specific prior written 00021 // permission. 00022 // 00023 // THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR 00024 // IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 00025 // OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN 00026 // NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 00027 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED 00028 // TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 00029 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 00030 // LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 00031 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 00032 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 00033 // 00034 // This file is part of the PLearn library. For more information on the PLearn 00035 // library, go to the PLearn Web site at www.plearn.org 00036 00037 00038 /* ******************************************************* 00039 * $Id: OneHotSquaredLoss.cc 4306 2005-10-23 02:42:13Z tihocan $ 00040 * This file is part of the PLearn library. 00041 ******************************************************* */ 00042 00043 #include "OneHotSquaredLoss.h" 00044 #include "RowAtPositionVariable.h" 00045 #include "Var_operators.h" 00046 //#include "Var_utils.h" 00047 00048 namespace PLearn { 00049 using namespace std; 00050 00051 00054 PLEARN_IMPLEMENT_OBJECT(OneHotSquaredLoss, 00055 "Computes sum(square_i(netout[i]-(i==classnum ?hotval :coldval))", 00056 "NO HELP"); 00057 00058 OneHotSquaredLoss::OneHotSquaredLoss() 00059 : coldval_(0.), hotval_(0.) 00060 { } 00061 00062 OneHotSquaredLoss::OneHotSquaredLoss(Variable* netout, Variable* classnum, real coldval, real hotval) 00063 : inherited(netout,classnum,1,1), coldval_(coldval), hotval_(hotval) 00064 { 00065 build_(); 00066 } 00067 00068 void 00069 OneHotSquaredLoss::build() 00070 { 00071 inherited::build(); 00072 build_(); 00073 } 00074 00075 void 00076 OneHotSquaredLoss::build_() 00077 { 00078 // input2 is classnum from constructor 00079 if(input2 && !input2->isScalar()) 00080 PLERROR("In OneHotSquaredLoss: classnum must be a scalar variable representing an index of netout (typically a classnum)"); 00081 } 00082 00083 void 00084 OneHotSquaredLoss::declareOptions(OptionList &ol) 00085 { 00086 declareOption(ol, "coldval_", &OneHotSquaredLoss::coldval_, OptionBase::buildoption, ""); 00087 declareOption(ol, "hotval_", &OneHotSquaredLoss::hotval_, OptionBase::buildoption, ""); 00088 inherited::declareOptions(ol); 00089 } 00090 00091 void OneHotSquaredLoss::recomputeSize(int& l, int& w) const 00092 { l=1, w=1; } 00093 00094 void OneHotSquaredLoss::fprop() 00095 { 00096 real* netout = input1->valuedata; 00097 int n = input1->value.size(); 00098 int classnum = (int) input2->valuedata[0]; 00099 real res = 0.; 00100 for(int i=0; i<n; i++) 00101 res += square(*netout++ - (i==classnum ? hotval_ : coldval_)); 00102 *valuedata = res; 00103 } 00104 00105 00106 void OneHotSquaredLoss::bprop() 00107 { 00108 real gr = *gradientdata; 00109 real* netout = input1->valuedata; 00110 int n = input1->value.size(); 00111 int classnum = (int) input2->valuedata[0]; 00112 real* input1grptr = input1->gradientdata; 00113 if(!fast_exact_is_equal(gr, 1.)) 00114 { 00115 gr = gr+gr; 00116 for(int i=0; i<n; i++) 00117 *input1grptr++ += gr*(*netout++ - (i==classnum ? hotval_ : coldval_)); 00118 } 00119 else // specialised version for gr==1 00120 { 00121 for(int i=0; i<n; i++) 00122 input1->gradientdata[i] += two(netout[i] - (i==classnum ? hotval_ : coldval_)); 00123 } 00124 } 00125 00126 00127 void OneHotSquaredLoss::symbolicBprop() 00128 { 00129 Var gi = g*(input1 - coldval_); 00130 Var gindex = new RowAtPositionVariable(g*(coldval_-hotval_), input2, input1->length()); 00131 Var ginput = gi + gindex; 00132 input1->accg(ginput+ginput); //2*gi 00133 } 00134 00135 00136 void OneHotSquaredLoss::rfprop() 00137 { 00138 int n = input1->value.size(); 00139 int classnum = (int) input2->valuedata[0]; 00140 real sum = 0; 00141 for (int i=0; i<n; i++) 00142 sum += 2 * input1->rvaluedata[i] * (input1->valuedata[i] - (i==classnum ? hotval_ : coldval_)); 00143 rvaluedata[0] = sum; 00144 } 00145 00146 00147 00148 } // end of namespace PLearn 00149 00150 00151 /* 00152 Local Variables: 00153 mode:c++ 00154 c-basic-offset:4 00155 c-file-style:"stroustrup" 00156 c-file-offsets:((innamespace . 0)(inline-open . 0)) 00157 indent-tabs-mode:nil 00158 fill-column:79 00159 End: 00160 */ 00161 // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=8:softtabstop=4:encoding=utf-8:textwidth=79 :