diff --git a/src/helper_files/helper.cpp b/src/helper_files/helper.cpp
index f0a4b68d859822934450f75626fdff0f323b1a70..57e3ffe7c71170be08998bf911e6e3c446b102e1 100644
--- a/src/helper_files/helper.cpp
+++ b/src/helper_files/helper.cpp
@@ -5,6 +5,8 @@
// Extra functions to reduce some code for the conversion between arma and float *array
+// arma --> array
+
float *convertToArray(colvec vec)
{
vector<float> vectorData = conv_to<vector<float>>::from(vec);
@@ -80,6 +82,9 @@ float *convertToArray(vector<vec> matrix)
return convertToArray(newVec);
}
+
+// array --> arma
+
rowvec convertArrayToRowvec(float *arr, int vecSize)
{
rowvec rVector(vecSize);
@@ -90,6 +95,16 @@ rowvec convertArrayToRowvec(float *arr, int vecSize)
return rVector;
}
+colvec convertArrayToColvec(float *arr, int vecSize)
+{
+ colvec rVector(vecSize);
+ for(int i = 0; i < vecSize; i++)
+ {
+ rVector[i] = arr[i];
+ }
+ return rVector;
+}
+
Row<size_t> convertArrayToVec(float *arr, int vecSize)
{
Row<size_t> rVector(vecSize);
@@ -110,6 +125,9 @@ mat convertArrayToMat(float *arr, int vecSize, int rowCount)
return matrix;
}
+
+
+
void returnMatrixInformation(mat matrix, float **mat, SP_integer *matColNum, SP_integer *matRowNum)
{
// return the Matrix dimensions
diff --git a/src/helper_files/helper.hpp b/src/helper_files/helper.hpp
index 7849b48a6d4373d91a2c41b88fabc0cfdbc2151f..f76ca68802a1969b8d8a18f71b7efb18b044de2b 100644
--- a/src/helper_files/helper.hpp
+++ b/src/helper_files/helper.hpp
@@ -31,6 +31,8 @@ float *convertToArray(vector<vec> vec);
rowvec convertArrayToRowvec(float *arr, int vecSize);
+colvec convertArrayToColvec(float *arr, int vecSize);
+
Row<size_t> convertArrayToVec(float *arr, int vecSize);
mat convertArrayToMat(float *arr, int vecSize, int rowCount);
diff --git a/src/methods/adaboost/adaboost.pl b/src/methods/adaboost/adaboost.pl
index a12de42f5ab70abe06ef61d4e85eb6246953a928..bd5df9ae189f00c38a26d978a05840895baa78e0 100644
--- a/src/methods/adaboost/adaboost.pl
+++ b/src/methods/adaboost/adaboost.pl
@@ -87,7 +87,7 @@ adaboost_classify(TestList, TestRows, PredicList, ProbsList, ZRows) :-
convert_float_array_to_list(Y, Ysize, PredicList),
convert_float_array_to_2d_list(Z, ZCols, ZRows, ProbsList).
-foreign(classify, c, classifyI( +pointer(float_array), +integer, +integer,
+foreign(classify, c, classifyI( +pointer(float_array), +integer, +integer,
-pointer(float_array), -integer,
-pointer(float_array), -integer, -integer)).
diff --git a/src/methods/linear_SVM/linear_SVM.cpp b/src/methods/linear_SVM/linear_SVM.cpp
index ae2994dc8d5a1fdd7f0720f902ffbf9d925bcea5..e7c31006913ecfad8eefcbdb3d5c1aa03000ca86 100644
--- a/src/methods/linear_SVM/linear_SVM.cpp
+++ b/src/methods/linear_SVM/linear_SVM.cpp
@@ -18,6 +18,7 @@ using namespace mlpack::svm;
// Global Variable of the LinearSVM object so it can be accessed from all functions
LinearSVM<> linearSVM;
+bool isModelTrained = false;
// input: const MatType & data,
@@ -39,6 +40,13 @@ void initModelWithTrain(float *dataMatArr, SP_integer dataMatSize, SP_integer da
{
// convert the Prolog array to arma::mat
mat data = convertArrayToMat(dataMatArr, dataMatSize, dataMatRowNum);
+ // check if labels fit the data
+ if (data.n_cols != labelsArrSize)
+ {
+ raisePrologSystemExeption("The number of data points does not match the number of labels!");
+ return;
+ }
+
// convert the Prolog array to arma::rowvec
arma::Row<size_t> labels = convertArrayToVec(labelsArr, labelsArrSize);
@@ -47,10 +55,12 @@ void initModelWithTrain(float *dataMatArr, SP_integer dataMatSize, SP_integer da
if (strcmp(optimizer, "lbfgs") == 0)
{
linearSVM.Train<ens::L_BFGS>(data, labels, numClasses);
+ isModelTrained = true;
}
else if (strcmp(optimizer, "psgd") == 0)
{
linearSVM.Train<ens::ParallelSGD<>>(data, labels, numClasses, ens::ParallelSGD(100, 4));
+ isModelTrained = true;
}
else
{
@@ -71,6 +81,8 @@ void initModelWithTrain(float *dataMatArr, SP_integer dataMatSize, SP_integer da
void initModelNoTrain(SP_integer numClasses, double lambda, double delta, SP_integer fitIntercept)
{
linearSVM = LinearSVM<>(numClasses, lambda, delta, true);
+
+ isModelTrained = false;
}
@@ -85,8 +97,20 @@ void classify(float *dataMatArr, SP_integer dataMatSize, SP_integer dataMatRowNu
float **labelsArr, SP_integer *labelsArrSize,
float **scoresMatArr, SP_integer *scoresMatColNum, SP_integer *scoresMatRowNum)
{
+ if (!isModelTrained)
+ {
+ raisePrologSystemExeption("The Model is not trained!");
+ return;
+ }
+
// convert the Prolog array to arma::mat
mat data = convertArrayToMat(dataMatArr, dataMatSize, dataMatRowNum);
+ // check if data has the correct feature size
+ if (data.n_rows != linearSVM.FeatureSize())
+ {
+ raisePrologSystemExeption("The given Datapoints Dimension is != the trained Datapoints Dimension!");
+ return;
+ }
// get the ReturnVector
arma::Row<size_t> labelsReturnVector;
@@ -119,12 +143,23 @@ void classify(float *dataMatArr, SP_integer dataMatSize, SP_integer dataMatRowNu
//
SP_integer classifyPoint(float *pointArr, SP_integer pointArrSize)
{
+ if (!isModelTrained)
+ {
+ raisePrologSystemExeption("The Model is not trained!");
+ return 0;
+ }
// convert the Prolog array to arma::rowvec
- rowvec pointVector = convertArrayToRowvec(pointArr, pointArrSize);
+ vec pointVector = convertArrayToColvec(pointArr, pointArrSize);
+ // check if data has the correct feature size
+ if (pointArrSize != linearSVM.FeatureSize())
+ {
+ raisePrologSystemExeption("The given Datapoint Dimension is != the trained Datapoints Dimension!");
+ return 0;
+ }
try
{
- return linearSVM.Classify<rowvec>(pointVector);
+ return linearSVM.Classify<vec>(pointVector);
}
catch(const std::exception& e)
{
@@ -143,8 +178,26 @@ SP_integer classifyPoint(float *pointArr, SP_integer pointArrSize)
//
double computeAccuracy(float *dataMatArr, SP_integer dataMatSize, SP_integer dataMatRowNum, float *labelsArr, SP_integer labelsArrSize)
{
+ if (!isModelTrained)
+ {
+ raisePrologSystemExeption("The Model is not trained!");
+ return 0.0;
+ }
// convert the Prolog array to arma::mat
mat data = convertArrayToMat(dataMatArr, dataMatSize, dataMatRowNum);
+ // check if data has the correct feature size
+ if (data.n_rows != linearSVM.FeatureSize())
+ {
+ raisePrologSystemExeption("The given Datapoints Dimension is != the trained Datapoints Dimension!");
+ return 0.0;
+ }
+ // check if labels fit the data
+ if (data.n_cols != labelsArrSize)
+ {
+ raisePrologSystemExeption("The number of data points does not match the number of labels!");
+ return 0.0;
+ }
+
// convert the Prolog array to arma::rowvec
arma::Row<size_t> labels = convertArrayToVec(labelsArr, labelsArrSize);
@@ -174,6 +227,13 @@ double train(float *dataMatArr, SP_integer dataMatSize, SP_integer dataMatRowNum
{
// convert the Prolog array to arma::mat
mat data = convertArrayToMat(dataMatArr, dataMatSize, dataMatRowNum);
+ // check if labels fit the data
+ if (data.n_cols != labelsArrSize)
+ {
+ raisePrologSystemExeption("The number of data points does not match the number of labels!");
+ return 0.0;
+ }
+
// convert the Prolog array to arma::rowvec
arma::Row<size_t> labels = convertArrayToVec(labelsArr, labelsArrSize);
@@ -181,10 +241,12 @@ double train(float *dataMatArr, SP_integer dataMatSize, SP_integer dataMatRowNum
{
if (strcmp(optimizer, "lbfgs") == 0)
{
+ isModelTrained = true;
return linearSVM.Train<ens::L_BFGS>(data, labels, numClasses);
}
else if (strcmp(optimizer, "psgd") == 0)
{
+ isModelTrained = true;
return linearSVM.Train<ens::ParallelSGD<>>(data, labels, numClasses, ens::ParallelSGD(100, 4));
}
else
diff --git a/src/methods/lmnn/lmnn.cpp b/src/methods/lmnn/lmnn.cpp
index b18b332fd429ba6f9f899c5e12ecbabf8dcf67cf..a818ebf2c2f6e6767151bd5bfcb38dd9f6e621b9 100644
--- a/src/methods/lmnn/lmnn.cpp
+++ b/src/methods/lmnn/lmnn.cpp
@@ -37,7 +37,12 @@ void lmnn(char const *optimizer,
{
// convert the Prolog arrays to arma::mat
mat data = convertArrayToMat(dataMatArr, dataMatSize, dataMatRowNum);
-
+ // check if labels fit the data
+ if (data.n_cols != labelsArrSize)
+ {
+ raisePrologSystemExeption("The number of data points does not match the number of labels!");
+ return;
+ }
// convert the Prolog arrays to arma::rowvec
Row< size_t > labelsVector = convertArrayToVec(labelsArr, labelsArrSize);
@@ -84,7 +89,7 @@ void lmnn(char const *optimizer,
lmnn.LearnDistance(outputReturnMat);
}
- else if (strcmp(optimizer, "amsgrad") == 0)
+ else if (strcmp(optimizer, "bbsgd") == 0)
{
LMNN<LMetric<2>, ens::BBS_BB> lmnn(data, labelsVector, k);
lmnn.Regularization() = regularization;
@@ -97,7 +102,7 @@ void lmnn(char const *optimizer,
lmnn.LearnDistance(outputReturnMat);
}
- else if (strcmp(optimizer, "amsgrad") == 0)
+ else if (strcmp(optimizer, "sgd") == 0)
{
// copied advise from the lmnn.main file
// Using SGD is not recommended as the learning matrix can
@@ -113,7 +118,7 @@ void lmnn(char const *optimizer,
lmnn.LearnDistance(outputReturnMat);
}
- else if (strcmp(optimizer, "amsgrad") == 0)
+ else if (strcmp(optimizer, "lbfgs") == 0)
{
LMNN<LMetric<2>, ens::L_BFGS> lmnn(data, labelsVector, k);
lmnn.Regularization() = regularization;