From: LaurentBerger <laurent.berger@univ-lemans.fr>
Date: Wed, 20 Dec 2017 14:24:46 +0000 (+0100)
Subject: Calcerror uses now weighted samples (#10346)
X-Git-Tag: accepted/tizen/6.0/unified/20201030.111113~265
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=e43997dbb56d5ce60adcde9dddb24b7f9f63922b;p=platform%2Fupstream%2Fopencv.git

Calcerror uses now weighted samples (#10346)

* Calcerror uses now sample weights

* catree comment in #10319
---

diff --git a/modules/ml/src/ann_mlp.cpp b/modules/ml/src/ann_mlp.cpp
index 5652da2..cd10c0d 100644
--- a/modules/ml/src/ann_mlp.cpp
+++ b/modules/ml/src/ann_mlp.cpp
@@ -155,7 +155,7 @@ int SimulatedAnnealingSolver::run()
             if (newEnergy < previousEnergy)
             {
                 previousEnergy = newEnergy;
-                //??? exchange++;
+                exchange++;
             }
             else
             {
@@ -405,21 +405,6 @@ public:
                 param2 = 0.1;
             params.bpMomentScale = std::min(param2, 1.);
         }
-/*        else if (method == ANN_MLP::ANNEAL)
-        {
-            if (param1 <= 0)
-                param1 = 10;
-            if (param2 <= 0 || param2>param1)
-                param2 = 0.1;
-            if (param3 <= 0 || param3 >=1)
-                param3 = 0.95;
-            if (param4 <= 0)
-                param4 = 10;
-            params.initialT = param1;
-            params.finalT = param2;
-            params.coolingRatio = param3;
-            params.itePerStep = param4;
-        }*/
     }
 
     int getTrainMethod() const
diff --git a/modules/ml/src/inner_functions.cpp b/modules/ml/src/inner_functions.cpp
index dc03e38..e95674a 100644
--- a/modules/ml/src/inner_functions.cpp
+++ b/modules/ml/src/inner_functions.cpp
@@ -94,26 +94,29 @@ public:
         int idxErr = range.start;
         CV_TRACE_FUNCTION_SKIP_NESTED();
         Mat samples = data->getSamples();
+        Mat weights=testerr? data->getTestSampleWeights() : data->getTrainSampleWeights();
         int layout = data->getLayout();
         Mat sidx = testerr ? data->getTestSampleIdx() : data->getTrainSampleIdx();
         const int* sidx_ptr = sidx.ptr<int>();
         bool isclassifier = s.isClassifier();
         Mat responses = data->getResponses();
         int responses_type = responses.type();
-
         double err = 0;
 
+
+        const float* sw = weights.empty() ? 0 : weights.ptr<float>();
         for (int i = range.start; i < range.end; i++)
         {
             int si = sidx_ptr ? sidx_ptr[i] : i;
+            double sweight = sw ? static_cast<double>(sw[i]) : 1.;
             Mat sample = layout == ROW_SAMPLE ? samples.row(si) : samples.col(si);
             float val = s.predict(sample);
             float val0 = (responses_type == CV_32S) ? (float)responses.at<int>(si) : responses.at<float>(si);
 
             if (isclassifier)
-                err += fabs(val - val0) > FLT_EPSILON;
+                err += sweight * fabs(val - val0) > FLT_EPSILON;
             else
-                err += (val - val0)*(val - val0);
+                err += sweight * (val - val0)*(val - val0);
             if (!resp.empty())
                 resp.at<float>(i) = val;
         }
@@ -133,12 +136,17 @@ float StatModel::calcError(const Ptr<TrainData>& data, bool testerr, OutputArray
     CV_TRACE_FUNCTION_SKIP_NESTED();
     Mat samples = data->getSamples();
     Mat sidx = testerr ? data->getTestSampleIdx() : data->getTrainSampleIdx();
+    Mat weights = testerr ? data->getTestSampleWeights() : data->getTrainSampleWeights();
     int n = (int)sidx.total();
     bool isclassifier = isClassifier();
     Mat responses = data->getResponses();
 
     if (n == 0)
+    {
         n = data->getNSamples();
+        weights = data->getTrainSampleWeights();
+        testerr =false;
+    }
 
     if (n == 0)
         return -FLT_MAX;
@@ -155,11 +163,11 @@ float StatModel::calcError(const Ptr<TrainData>& data, bool testerr, OutputArray
 
     for (size_t i = 0; i < errStrip.size(); i++)
         err += errStrip[i];
-
+    float weightSum= weights.empty() ? n: static_cast<float>(sum(weights)(0));
     if (_resp.needed())
         resp.copyTo(_resp);
 
-    return (float)(err / n * (isclassifier ? 100 : 1));
+    return (float)(err/ weightSum * (isclassifier ? 100 : 1));
 }
 
 /* Calculates upper triangular matrix S, where A is a symmetrical matrix A=S'*S */