void checkDataRegression(double* values, LabeledData<RealVector,RealVector> const& loaded, std::size_t labelStart, std::size_t labelEnd){
BOOST_REQUIRE_EQUAL(loaded.numberOfElements(),numInputs);
std::size_t inputStart =0;
std::size_t inputEnd = labelStart;
if(labelStart == 0){
inputStart = labelEnd;
inputEnd = numDimensions+1;
}
for (size_t i=0; i != numInputs; ++i){
for (size_t j=0; j != numDimensions+1; ++j)
{
double element = 0;
if(j >= labelStart &&j < labelEnd){
element = loaded.element(i).label(j-labelStart);
}
if(j >= inputStart && j < inputEnd){
element = loaded.element(i).input(j-inputStart);
}
if( boost::math::isnan(values[i*(numDimensions+1)+j])){
BOOST_CHECK(boost::math::isnan(element));
}
else
{
BOOST_CHECK_EQUAL(element, values[i*(numDimensions+1)+j]);
}
}
}
}
TEST(Csv, ReadCsv) {
LabeledData<RealVector, uint32_t> data;
EXPECT_EQ(0U, data.size());
EXPECT_FALSE(ReadCsv("null/null", &data));
ASSERT_TRUE(ReadCsv("testdata/data/cls.10.csv", &data, LAST_COLUMN));
const std::size_t kN = 10;
EXPECT_EQ(kN, data.size());
uint32_t expected_labels_arr[kN] = {0, 1, 0, 1, 0, 1, 0, 1, 0, 1};
Data<uint32_t> expected_labels;
ToData(expected_labels_arr, kN, &expected_labels);
EXPECT_EQ(expected_labels, data.labels());
EXPECT_EQ(2U, data.input(0).size());
EXPECT_DOUBLE_EQ(2.4114, data.input(0)(0));
EXPECT_DOUBLE_EQ(-3.8901, data.input(0)(1));
ASSERT_TRUE(ReadCsv("testdata/data/cls.csv", &data));
EXPECT_EQ(1000U, data.size());
LabeledData<RealVector, double> data2;
ASSERT_TRUE(ReadCsv("testdata/data/cls.10.csv", &data2, FIRST_COLUMN));
EXPECT_EQ(10U, data2.size());
EXPECT_DOUBLE_EQ(2.4114, data2.label(0));
EXPECT_EQ(2U, data2.input(0).size());
EXPECT_DOUBLE_EQ(-3.8901, data2.input(0)(0));
EXPECT_DOUBLE_EQ(0, data2.input(0)(1));
}
void LinearRegression::train(LinearModel<>& model, LabeledData<RealVector, RealVector> const& dataset){
std::size_t inputDim = inputDimension(dataset);
std::size_t outputDim = labelDimension(dataset);
std::size_t numInputs = dataset.numberOfElements();
std::size_t numBatches = dataset.numberOfBatches();
//Let P be the matrix of points with n rows and X=(P|1). the 1 rpresents the bias weight
//Let A = X^T X + lambda * I
//than whe have (for lambda = 0)
//A = ( P^T P P^T 1)
// ( 1^T P 1^T1)
RealMatrix matA(inputDim+1,inputDim+1,0.0);
blas::Blocking<RealMatrix> Ablocks(matA,inputDim,inputDim);
//compute A and the label matrix batchwise
typedef LabeledData<RealVector, RealVector>::const_batch_reference BatchRef;
for (std::size_t b=0; b != numBatches; b++){
BatchRef batch = dataset.batch(b);
symm_prod(trans(batch.input),Ablocks.upperLeft(),false);
noalias(column(Ablocks.upperRight(),0))+=sum_rows(batch.input);
}
row(Ablocks.lowerLeft(),0) = column(Ablocks.upperRight(),0);
matA(inputDim,inputDim) = numInputs;
//X^TX+=lambda* I
diag(Ablocks.upperLeft())+= blas::repeat(m_regularization,inputDim);
//we also need to compute X^T L= (P^TL, 1^T L) where L is the matrix of labels
RealMatrix XTL(inputDim + 1,outputDim,0.0);
for (std::size_t b=0; b != numBatches; b++){
BatchRef batch = dataset.batch(b);
RealSubMatrix PTL = subrange(XTL,0,inputDim,0,outputDim);
axpy_prod(trans(batch.input),batch.label,PTL,false);
noalias(row(XTL,inputDim))+=sum_rows(batch.label);
}
//we solve the system A Beta = X^T L
//usually this is solved via the moore penrose inverse:
//Beta = A^-1 T
//but it is faster und numerically more stable, if we solve it as a symmetric system
//w can use in-place solve
RealMatrix& beta = XTL;
blas::solveSymmSemiDefiniteSystemInPlace<blas::SolveAXB>(matA,beta);
RealMatrix matrix = subrange(trans(beta), 0, outputDim, 0, inputDim);
RealVector offset = row(beta,inputDim);
// write parameters into the model
model.setStructure(matrix, offset);
}
void CMySharkML::Features2SharkData(LabeledData<RealVector, unsigned int> &dataset, cv::Mat &features, std::vector<int> &v_label)
{
//copy rows of the file into the dataset
std::size_t rows = features.rows;
std::size_t dimensions = features.cols;
std::vector<std::size_t> batchSizes = shark::detail::optimalBatchSizes(rows, 256);
// Test data
dataset = LabeledData<RealVector, unsigned int>(batchSizes.size());
std::size_t currentRow = 0;
for(std::size_t b = 0; b != batchSizes.size(); ++b) {
RealMatrix& inputs = dataset.batch(b).input;
UIntVector& labels = dataset.batch(b).label;
inputs.resize(batchSizes[b], dimensions);
labels.resize(batchSizes[b]);
//copy the rows into the batch
for(std::size_t i = 0; i != batchSizes[b]; ++i,++currentRow){
int rawLabel = v_label[currentRow];
labels[i] = rawLabel;
for(std::size_t j = 0; j != dimensions; ++j){
inputs(i,j) = features.at<float>(currentRow, j);
}
}
}
}
void checkDataEquality(T* values, unsigned int* labels, LabeledData<V,U> const& loaded){
BOOST_REQUIRE_EQUAL(loaded.numberOfElements(),numInputs);
BOOST_REQUIRE_EQUAL(inputDimension(loaded),numDimensions);
for (size_t i=0; i != numInputs; ++i){
for (size_t j=0; j != numDimensions; ++j)
{
if( boost::math::isnan(values[i*numDimensions+j])){
BOOST_CHECK(boost::math::isnan(loaded.element(i).input(j)));
}
else
{
BOOST_CHECK_EQUAL(loaded.element(i).input(j), values[i*numDimensions+j]);
}
}
BOOST_CHECK_EQUAL(loaded.element(i).label, labels[i]);
}
}
/// append new data to the training data
void TrainingData::appendData(const LabeledData &labeledData)
{
assert((labeledData.getNumPosExamples() + labeledData.getNumNegExamples()) == labeledData.getNumExamples());
const size_t
initialNumberOfTrainingSamples = getNumExamples(),
finalNumberOfTrainingSamples = initialNumberOfTrainingSamples + labeledData.getNumExamples();
if(finalNumberOfTrainingSamples > getMaxNumExamples())
{
throw std::runtime_error("TrainingData::appendData is trying to add more data than initially specified");
}
// compute and save the feature responses
#pragma omp parallel for default(none) ordered shared(labeledData)
for (size_t labelDataIndex = 0; labelDataIndex < labeledData.getNumExamples(); ++labelDataIndex)
{
const meta_datum_t &metaDatum = labeledData.getMetaDatum(labelDataIndex);
const LabeledData::integral_channels_t &integralImage = labeledData.getIntegralImage(labelDataIndex);
setDatum(initialNumberOfTrainingSamples + labelDataIndex,
metaDatum, integralImage);
} // end of "for each labeled datum"
return;
}
int main(){
//get problem data
Problem problem(1.0);
LabeledData<RealVector,unsigned int> training = problem.generateDataset(1000);
LabeledData<RealVector,unsigned int> test = problem.generateDataset(100);
std::size_t inputs=inputDimension(training);
std::size_t outputs = numberOfClasses(training);
std::size_t hiddens = 10;
unsigned numberOfSteps = 1000;
//create network and initialize weights random uniform
FFNet<LogisticNeuron,LinearNeuron> network;
network.setStructure(inputs,hiddens,outputs);
initRandomUniform(network,-0.1,0.1);
//create error function
CrossEntropy loss;
ErrorFunction error(training,&network,&loss);
// loss for evaluation
// The zeroOneLoss for multiclass problems assigns the class to the highest output
ZeroOneLoss<unsigned int, RealVector> loss01;
// evaluate initial network
Data<RealVector> prediction = network(training.inputs());
cout << "classification error before learning:\t" << loss01.eval(training.labels(), prediction) << endl;
//initialize Rprop
IRpropPlus optimizer;
optimizer.init(error);
for(unsigned step = 0; step != numberOfSteps; ++step)
optimizer.step(error);
// evaluate solution found by training
network.setParameterVector(optimizer.solution().point); // set weights to weights found by learning
prediction = network(training.inputs());
cout << "classification error after learning:\t" << loss01(training.labels(), prediction) << endl;
}
void LDA::train(LinearClassifier<>& model, LabeledData<RealVector,unsigned int> const& dataset){
if(dataset.empty()){
throw SHARKEXCEPTION("[LDA::train] the dataset must not be empty");
}
std::size_t inputs = dataset.numberOfElements();
std::size_t dim = inputDimension(dataset);
std::size_t classes = numberOfClasses(dataset);
//required statistics
UIntVector num(classes,0);
RealMatrix means(classes, dim,0.0);
RealMatrix covariance(dim, dim,0.0);
//we compute the data batch wise
for(auto const& batch: dataset.batches()){
UIntVector const& labels = batch.label;
RealMatrix const& points = batch.input;
//load batch and update mean
std::size_t currentBatchSize = points.size1();
for (std::size_t e=0; e != currentBatchSize; e++){
//update mean and class count for this sample
std::size_t c = labels(e);
++num(c);
noalias(row(means,c))+=row(points,e);
}
//update second moment matrix
noalias(covariance) += prod(trans(points),points);
}
covariance/=inputs-classes;
//calculate mean and the covariance matrix from second moment
for (std::size_t c = 0; c != classes; c++){
if (num[c] == 0)
throw SHARKEXCEPTION("[LDA::train] LDA can not handle a class without examples");
row(means,c) /= num(c);
double factor = num(c);
factor/=inputs-classes;
noalias(covariance)-= factor*outer_prod(row(means,c),row(means,c));
}
//add regularization
if(m_regularization>0){
for(std::size_t i=0;i!=dim;++i)
covariance(i,i)+=m_regularization;
}
//the formula for the linear classifier is
// arg max_i log(P(x|i) * P(i))
//= arg max_i log(P(x|i)) +log(P(i))
//= arg max_i -(x-m_i)^T C^-1 (x-m_i) +log(P(i))
//= arg max_i -m_i^T C^-1 m_i +2* x^T C^-1 m_i + log(P(i))
//so we compute first C^-1 m_i and then the first term
// compute z = m_i^T C^-1 <=> z C = m_i
// this is the expensive step of the calculation.
RealMatrix transformedMeans = means;
blas::solveSymmSemiDefiniteSystemInPlace<blas::SolveXAB>(covariance,transformedMeans);
//compute bias terms m_i^T C^-1 m_i - log(P(i))
RealVector bias(classes);
for(std::size_t c = 0; c != classes; ++c){
double prior = std::log(double(num(c))/inputs);
bias(c) = - 0.5* inner_prod(row(means,c),row(transformedMeans,c)) + prior;
}
//fill the model
model.decisionFunction().setStructure(transformedMeans,bias);
}
// optimize the sigmoid using platt's method
void SigmoidFitPlatt::train(SigmoidModel& model, LabeledData<RealVector, unsigned int> const& dataset){
SIZE_CHECK( model.numberOfParameters() == 2 );
typedef LabeledData<RealVector, unsigned int>::const_element_range Elements;
typedef IndexedIterator<boost::range_iterator<Elements>::type> Iterator;
double a, b, c, d, e, d1, d2;
double t = 0.0;
double oldA, oldB, diff, scale, det;
double err = 0.0;
double value, p;
double lambda = 0.001;
double olderr = 1e100;
std::vector<std::size_t> classCount = classSizes(dataset);
std::size_t pos = classCount[1];
std::size_t neg = classCount[0];
std::size_t ic = pos+neg;
Iterator end(dataset.elements().end(),ic);
double A = 0.0;
double B = std::log((neg + 1.0) / (pos + 1.0));
double lowTarget = 1.0 / (neg + 2.0);
double highTarget = (pos + 1.0) / (pos + 2.0);
RealVector pp(ic,(pos + 1.0) / (pos + neg + 2.0));
int count = 0;
for (std::size_t it = 0; it < 100; it++)
{
a = b = c = d = e = 0.0;
for (Iterator it(dataset.elements().begin(),0); it != end; ++it){
std::size_t i = it.index();
t = (it->label == 1) ? highTarget : lowTarget;
d1 = pp(i) - t;
d2 = pp(i) * (1.0 - pp(i));
value = it->input(0);
a += d2 * value * value;
b += d2;
c += d2 * value;
d += d1 * value;
e += d1;
}
if (std::abs(d) < 1e-9 && std::abs(e) < 1e-9) break;
oldA = A;
oldB = B;
err = 0.0;
while (true)
{
det = (a + lambda) * (b + lambda) - c * c;
if (det == 0.0)
{
lambda *= 10.0;
continue;
}
A = oldA + ((b + lambda) * d - c * e) / det;
B = oldB + ((a + lambda) * e - c * d) / det;
err = 0.0;
for (Iterator it(dataset.elements().begin(),0); it != end; ++it){
std::size_t i = it.index();
p = 1.0 / (1.0 + std::exp(A * it->input(0) + B));
pp(i) = p;
err -= t * safeLog(p) + (1.0 - t) * safeLog(1.0 - p);
}
if (err < 1.0000001 * olderr)
{
lambda *= 0.1;
break;
}
lambda *= 10.0;
if (lambda >= 1e6)
{
// Something is broken. Give up.
break;
}
}
diff = err - olderr;
scale = 0.5 * (err + olderr + 1.0);
if (diff > -1e-3 * scale && diff < 1e-7 * scale)
count++;
else
count = 0;
olderr = err;
if (count == 3) break;
}
RealVector params(2);
params(0) = A;
params(1) = B;
model.setParameterVector(params);
}
void testDatasetEquality(LabeledData<int, int> const& set1, LabeledData<int, int> const& set2){
BOOST_REQUIRE_EQUAL(set1.numberOfBatches(),set2.numberOfBatches());
BOOST_REQUIRE_EQUAL(set1.numberOfElements(),set2.numberOfElements());
for(std::size_t i = 0; i != set1.numberOfBatches(); ++i){
BOOST_REQUIRE_EQUAL(set1.batch(i).input.size(),set1.batch(i).label.size());
BOOST_REQUIRE_EQUAL(set2.batch(i).input.size(),set2.batch(i).label.size());
}
testSetEquality(set1.inputs(),set2.inputs());
testSetEquality(set1.labels(),set2.labels());
}
