Merge branch 'ProtoNN-dev' of https://github.com/Microsoft/EdgeML into ProtoNN-dev

BonsaiLocalDriver.cpp

@@ -38,7 +38,7 @@ int main(int argc, char **argv)
   BonsaiPredictor predictor(modelBytes, model); // use the constructor predictor(modelBytes, model, false) for loading a sparse model.
   predictor.importMeanVar(meanVarBytes, meanVar);
   
-  predictor.batchEvaluate(trainer.data.Xtest, trainer.data.Ytest, dataDir, currResultsPath);
+  predictor.batchEvaluate(trainer.data.Xvalidation, trainer.data.Yvalidation, dataDir, currResultsPath);
   
   delete[] model, meanVar;
   

src/Bonsai/Bonsai.h

@@ -80,7 +80,7 @@ namespace EdgeML
 
         int seed;
         int iters, epochs;
-        dataCount_t ntrain, ntest, batchSize;
+        dataCount_t ntrain, nvalidation, batchSize;
         bool isOneIndex;
 
         FP_TYPE Sigma; ///< Sigmoid parameter for prediction

src/Bonsai/BonsaiFunctions.cpp

@@ -957,7 +957,7 @@ void Bonsai::parseInput(const int& argc, const char** argv,
 		required++;
 		break;
 	  case 'E':
-		hyperParam.ntest = int(atoi(argv[i]));
+		hyperParam.nvalidation = int(atoi(argv[i]));
 		required++;
 		break;
 	  }

src/Bonsai/BonsaiHyperParams.cpp

@@ -20,7 +20,7 @@ BonsaiModel::BonsaiHyperParams::BonsaiHyperParams()
   seed = 42;
 
   ntrain = 0;
-  ntest = 0;
+  nvalidation = 0;
   batchSize = 0;
 
   iters = 0;
@@ -102,7 +102,6 @@ void BonsaiModel::BonsaiHyperParams::finalizeHyperParams()
   // Following asserts removed to faciliate support for TLC
   // which does not know how many datapoints are going to be fed before-hand!
   // assert(ntrain >= 1);               
-  // assert(ntest >= 0);
   assert(projectionDimension <= dataDimension + 1);
   assert(numClasses > 0);
 

src/Bonsai/BonsaiIngestTest.cpp

@@ -24,7 +24,7 @@ int main()
 
   hyperParam.numClasses = 10;
 
-  hyperParam.ntest = 0;
+  hyperParam.nvalidation = 0;
   hyperParam.ntrain = 5000;
 
   hyperParam.Sigma = 1.0;

src/Bonsai/BonsaiTrainer.cpp

@@ -29,7 +29,7 @@ BonsaiTrainer::BonsaiTrainer(
   data(dataIngestType,
     DataFormatParams{
       model.hyperParams.ntrain,
-        model.hyperParams.ntest,
+        model.hyperParams.nvalidation,
         model.hyperParams.numClasses,
         model.hyperParams.dataDimension })
 {
@@ -47,7 +47,7 @@ BonsaiTrainer::BonsaiTrainer(
   mean = MatrixXuf::Zero(model.hyperParams.dataDimension, 1);
   variance = MatrixXuf::Zero(model.hyperParams.dataDimension, 1);
 
-  data.loadDataFromFile(model.hyperParams.dataformatType, dataDir + "/train.txt", dataDir + "/test.txt");
+  data.loadDataFromFile(model.hyperParams.dataformatType, dataDir + "/train.txt", dataDir + "/test.txt", "");
   finalizeData();
 }
 
@@ -62,7 +62,7 @@ BonsaiTrainer::BonsaiTrainer(
   data(dataIngestType,
     DataFormatParams{
       model.hyperParams.ntrain,
-  model.hyperParams.ntest,
+  model.hyperParams.nvalidation,
   model.hyperParams.numClasses,
   model.hyperParams.dataDimension })
 {
@@ -81,7 +81,7 @@ BonsaiTrainer::BonsaiTrainer(
   mean = MatrixXuf::Zero(model.hyperParams.dataDimension, 1);
   variance = MatrixXuf::Zero(model.hyperParams.dataDimension, 1);
 
-  data.loadDataFromFile(model.hyperParams.dataformatType, dataDir + "/train.txt", dataDir + "/test.txt");
+  data.loadDataFromFile(model.hyperParams.dataformatType, dataDir + "/train.txt", dataDir + "/test.txt", "");
   finalizeData();
 
   initializeModel();
@@ -96,7 +96,7 @@ BonsaiTrainer::BonsaiTrainer(
   data(dataIngestType,
     DataFormatParams{
          model.hyperParams.ntrain,
-     model.hyperParams.ntest,
+     model.hyperParams.nvalidation,
      model.hyperParams.numClasses,
      model.hyperParams.dataDimension })
 {
@@ -155,12 +155,11 @@ void BonsaiTrainer::finalizeData()
     // This condition means that the ingest type is Interface ingest,
     // hence the number of training points was not known beforehand. 
     model.hyperParams.ntrain = data.Xtrain.cols();
-    assert(data.Xtest.cols() == 0);
-    model.hyperParams.ntest = 0;
+    assert(data.Xvalidation.cols() == 0);
+    model.hyperParams.nvalidation = 0;
   }
   else {
     assert(model.hyperParams.ntrain == data.Xtrain.cols());
-    // assert(model.hyperParams.ntest == data.Xtest.cols());
   }
 
   // Following asserts can only be made in finalieData since TLC 
@@ -340,11 +339,15 @@ void BonsaiTrainer::exportMeanVar(
 void BonsaiTrainer::normalize()
 {
   if (model.hyperParams.normalizationType == minMax) {
-    minMaxNormalize(data.Xtrain, data.Xtest);
+    computeMinMax(data.Xtrain, data.min, data.max);
+    minMaxNormalize(data.Xtrain, data.min, data.max);
+    if (data.Xvalidation.cols() > 0)
+      minMaxNormalize(data.Xvalidation, data.min, data.max);
   }
   else if (model.hyperParams.normalizationType == l2) {
     l2Normalize(data.Xtrain);
-    l2Normalize(data.Xtest);
+    if (data.Xvalidation.cols() > 0)
+      l2Normalize(data.Xvalidation);
   }
   else;
 }