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: NegLogPoissonVariable.cc 4412 2005-11-02 19:00:20Z tihocan $ 00040 * This file is part of the PLearn library. 00041 ******************************************************* */ 00042 00043 #include "NegLogPoissonVariable.h" 00044 00045 namespace PLearn { 00046 using namespace std; 00047 00050 PLEARN_IMPLEMENT_OBJECT(NegLogPoissonVariable, 00051 "Negative loglikelihood of the poisson distribution", 00052 "Negative loglikelihood of the poisson distribution\n" 00053 "cost_i = {exp(output_i) - output_i * target_i + log(target_i!)} / weight_i\n" 00054 "cost = sum_i weight_i * cost_i\n" 00055 "Thus, wrt usual notation, output_i = log(lambda_i)\n" 00056 "Note that, in order to comply with regular procedures, cost_i\n" 00057 "is divided by weight_i and multiplied back when computing the weighted cost"); 00058 00059 // We can link the notation used above to the usual 00060 // statistical notation in the following way: 00061 // lambda = exp(output_i) 00062 // where exponentiation ensures a positive value for lambda 00063 // and k = target_i, the number of observed events 00064 // Thus, the cost corresponds to the negative log likelihood 00065 // of the Poisson density 00066 00067 NegLogPoissonVariable::NegLogPoissonVariable(VarArray& the_varray) 00068 : inherited(the_varray, 1, 1) 00069 { 00070 build_(); 00071 } 00072 00073 void 00074 NegLogPoissonVariable::build() 00075 { 00076 inherited::build(); 00077 build_(); 00078 } 00079 00080 void 00081 NegLogPoissonVariable::build_() 00082 { 00083 if (varray[0] && varray[1] && (varray[0]->size() != varray[1]->size())) 00084 PLERROR("In NegLogPoissonVariable: netout and target must have the same size"); 00085 if (varray.size()>2 && varray[0] && varray[2] && (varray[0]->size() != varray[2]->size())) 00086 PLERROR("In NegLogPoissonVariable: netout and weight must have the same size"); 00087 } 00088 00089 void NegLogPoissonVariable::recomputeSize(int& l, int& w) const 00090 { l=1, w=1; } 00091 00092 void NegLogPoissonVariable::fprop() 00093 { 00094 real cost = 0.0; 00095 for (int i=0; i<varray[0]->size(); i++) 00096 { 00097 real output = varray[0]->valuedata[i]; 00098 real target = varray[1]->valuedata[i]; 00099 real weight = 1; 00100 if (varray.size()>2) 00101 weight = varray[2]->valuedata[i]; 00102 00103 static real EPSILON = 1e-8; // Regularization 00104 real log_fact_target = pl_gammln(max(target, EPSILON)+1); 00105 // cost += exp(output) * weight - (output + pl_log(weight) ) * target + log_fact_target; 00106 00107 if (weight > 0) 00108 cost += exp(output) - (output + pl_log(weight)) * target / weight + 00109 log_fact_target / weight; 00110 } 00111 if (is_missing(cost)) 00112 PLERROR("NegLogPoissonVariable::fprop: encountered NaN cost"); 00113 00114 valuedata[0] = cost; 00115 } 00116 00117 void NegLogPoissonVariable::bprop() 00118 { 00119 real gr = gradient[0]; 00120 for (int i=0; i<varray[0]->size(); i++) 00121 { 00122 real output = varray[0]->valuedata[i]; 00123 real target = varray[1]->valuedata[i]; 00124 real weight = 1; 00125 if (varray.size()>2) 00126 weight = varray[2]->valuedata[i]; 00127 00128 if (weight > 0) 00129 varray[0]->gradientdata[i] += gr * ( exp(output) - target / weight); 00130 // varray[0]->gradientdata[i] += gr* ( exp(output) * weight - target); 00131 } 00132 } 00133 00134 } // end of namespace PLearn 00135 00136 00137 /* 00138 Local Variables: 00139 mode:c++ 00140 c-basic-offset:4 00141 c-file-style:"stroustrup" 00142 c-file-offsets:((innamespace . 0)(inline-open . 0)) 00143 indent-tabs-mode:nil 00144 fill-column:79 00145 End: 00146 */ 00147 // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=8:softtabstop=4:encoding=utf-8:textwidth=79 :