From 7a55173001827c68a9cc7babc689c7c17844e7d7 Mon Sep 17 00:00:00 2001
From: Thomas Purcell <purcell@fhi-berlin.mpg.de>
Date: Wed, 23 Sep 2020 11:19:45 +0200
Subject: [PATCH] Update SVM to use libsvm not shark ML
Something is wrong with sharkML in some enviroments
---
CMakeLists.txt | 13 +-
.../Model/ModelClassifier.cpp | 309 ++++++++++--------
.../Model/ModelClassifier.hpp | 10 +-
src/external/libsvm/svm.cpp | 46 +--
src/main.cpp | 2 -
.../descriptor_identifier/SISSOClassifier.cpp | 22 +-
6 files changed, 224 insertions(+), 178 deletions(-)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0ca95102..277813c9 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -21,7 +21,7 @@ set(CMAKE_CXX_EXTENSIONS OFF)
# compiler options
option(EXTERNAL_BOOST "Use an external boost library" OFF)
-option(USE_PYTHON "Whether to compile with python binding support" OFF)
+option(USE_PYTHON "Whether to compile with python binding support" ON)
if(USE_PYTHON)
message(STATUS "USE PYTHON True")
@@ -277,7 +277,16 @@ set(SHARK_BUILD_DIR "${CMAKE_CURRENT_BINARY_DIR}/external/shark/build/")
set(SHARK_INSTALL_DIR "${CMAKE_CURRENT_BINARY_DIR}/external/shark/bin/")
set(SHARK_INCLUDE_DIRS "${SHARK_INSTALL_DIR}/include")
set(SHARK_LIBRARY_DIRS "${SHARK_INSTALL_DIR}/lib")
-set(SHARK_CMAKE_ARGS "CXX=${CMAKE_CXX_COMPILER};-DBUILD_EXAMPLES=OFF;-DBUILD_TESTING=OFF;-DBUILD_SHARED_LIBS=ON;-DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER};-DCMAKE_INSTALL_PREFIX=${SHARK_INSTALL_DIR};-DCBLAS_LIBRARY_PATH=${LAPACK_LIBRARY_DIR};-DBOOST_ROOT=${Boost_INSTALL_DIR};-DCMAKE_INSTALL_LIBDIR=${SHARK_LIBRARY_DIRS}")
+set(SHARK_CMAKE_ARGS "CXX=${CMAKE_CXX_COMPILER}")
+list(APPEND SHARK_CMAKE_ARGS "-DBUILD_EXAMPLES=OFF")
+list(APPEND SHARK_CMAKE_ARGS "-DBUILD_TESTING=OFF")
+list(APPEND SHARK_CMAKE_ARGS "-DBUILD_SHARED_LIBS=ON")
+list(APPEND SHARK_CMAKE_ARGS "-DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}")
+list(APPEND SHARK_CMAKE_ARGS "-DCMAKE_INSTALL_PREFIX=${SHARK_INSTALL_DIR}")
+list(APPEND SHARK_CMAKE_ARGS -DCBLAS_LIBRARY_PATH=${LAPACK_LIBRARY_DIR})
+list(APPEND SHARK_CMAKE_ARGS -DBOOST_ROOT=${Boost_INSTALL_DIR})
+list(APPEND SHARK_CMAKE_ARGS -DCMAKE_INSTALL_LIBDIR=${SHARK_LIBRARY_DIRS})
+
ExternalProject_Add(
external_shark
PREFIX "external/shark"
diff --git a/src/descriptor_identifier/Model/ModelClassifier.cpp b/src/descriptor_identifier/Model/ModelClassifier.cpp
index 5056ebf5..b3a3c8f7 100644
--- a/src/descriptor_identifier/Model/ModelClassifier.cpp
+++ b/src/descriptor_identifier/Model/ModelClassifier.cpp
@@ -135,152 +135,197 @@ ModelClassifier::ModelClassifier(Unit prop_unit, std::vector<double> prop_train,
std::transform(_prop_train_est.begin(), _prop_train_est.end(), _prop_train.begin(), _train_error.begin(), [](double est, double real){return (std::abs(est - real) < 1e-10) ? -1 : real;});
std::transform(_prop_test_est.begin(), _prop_test_est.end(), _prop_test.begin(), _test_error.begin(), [](double est, double real){return (std::abs(est - real) < 1e-10) ? -1 : real;});
- // Setup ML Shark data structures for SVM training
- std::vector<shark::LabeledData<shark::RealVector, unsigned int>> training(_task_sizes_train.size());
- std::vector<shark::LabeledData<shark::RealVector, unsigned int>> test(_task_sizes_test.size());
- std::vector<unsigned int> prop_to_copy(_prop_train.size());
- int n_els = 0;
-
- // Populate data structures for SVM and convex hull overlap
+ // Perform SVM
+ int start = 0;
+ int start_test = 0;
for(int tt = 0; tt < _task_sizes_train.size(); ++tt)
{
- shark::Data<shark::RealVector> data(_task_sizes_train[tt], shark::RealVector(_n_dim));
- shark::Data<unsigned int> labels(_task_sizes_train[tt], 0);
-
- for(int ii = 0; ii < _task_sizes_train[tt]; ++ii)
- prop_to_copy[ii] = static_cast<unsigned int>(_prop_train[ii + n_els]);
+ SVMWrapper svm(n_class, _n_dim, _task_sizes_train[tt], &_prop_train[start]);
- int n_copied = 0;
- for(auto& batch : labels.batches())
- {
- std::copy_n(prop_to_copy.begin() + n_copied, shark::batchSize(batch), &batch(0));
- n_els += shark::batchSize(batch);
- n_copied += shark::batchSize(batch);
- }
+ std::vector<double*> node_val_ptrs(_n_dim);
+ std::vector<double*> node_test_val_ptrs(_n_dim);
- n_copied = 0;
- for(auto& batch : data.batches())
+ for(int dd = 0; dd < _n_dim; ++dd)
{
- for(int ff = 0; ff < _n_dim; ++ff)
- dcopy_(shark::batchSize(batch), _feats[ff]->svm_value_ptr() + n_copied, 1, &batch(0, 0) + ff, _n_dim);
-
- n_copied += shark::batchSize(batch);
+ node_val_ptrs[dd] = _feats[dd]->value_ptr() + start;
+ node_test_val_ptrs[dd] = _feats[dd]->test_value_ptr() + start;
}
- training[tt] = shark::LabeledData<shark::RealVector, unsigned int>(data, labels);
- }
-
- prop_to_copy.resize(_prop_test.size());
- n_els = 0;
-
- for(int tt = 0; tt < _task_sizes_test.size(); ++tt)
- {
- if(_task_sizes_test[tt] == 0)
- continue;
-
- shark::Data<shark::RealVector> data(_task_sizes_test[tt], shark::RealVector(_n_dim));
- shark::Data<unsigned int> labels(_task_sizes_test[tt], 0);
- for(int ii = 0; ii < _task_sizes_test[tt]; ++ii)
- prop_to_copy[ii] = static_cast<unsigned int>(_prop_test[ii + n_els]);
+ svm.train(node_val_ptrs);
- int n_copied = 0;
- for(auto& batch : labels.batches())
+ int n_coef = 0;
+ for(auto& coef : svm.coefs())
{
- std::copy_n(prop_to_copy.begin() + n_copied, shark::batchSize(batch), &batch(0));
- n_els += shark::batchSize(batch);
- n_copied += shark::batchSize(batch);
+ _coefs.push_back(coef);
+ _coefs.back().push_back(svm.intercept()[n_coef]);
+ // for(int dd = 0; dd < _n_dim; ++dd)
+ // {
+ // _coefs.back()[dd] = _feats[dd]->remap_w_svm(_coefs.back()[dd]);
+ // _coefs.back()[_n_dim] -= _feats[dd]->remap_b_svm(_coefs.back()[dd]);
+ // }
+ ++n_coef;
}
- n_copied = 0;
- for(auto& batch : data.batches())
- {
- for(int ff = 0; ff < _n_dim; ++ff)
- dcopy_(shark::batchSize(batch), _feats[ff]->svm_test_value_ptr() + n_copied, 1, &batch(0, 0) + ff, _n_dim);
+ std::copy_n(svm.y_estimate().begin() + start, _task_sizes_train[tt], _prop_train_est.begin() + start);
+ std::copy_n(svm.predict(_task_sizes_test[tt], node_test_val_ptrs).begin() + start_test, _task_sizes_test[tt], _prop_test_est.begin() + start_test);
- n_copied += shark::batchSize(batch);
- }
- test[tt] = shark::LabeledData<shark::RealVector, unsigned int>(data, labels);
+ start += _task_sizes_train[tt];
+ start_test += _task_sizes_test[tt];
}
- // Perform SVM training
-
- // Process SVM model
- std::shared_ptr<shark::AbstractKernelFunction<shark::RealVector>> _kernel_ptr = std::make_shared<shark::LinearKernel<shark::RealVector>>();
- shark::CSvmTrainer<shark::RealVector> trainer(_kernel_ptr.get(), 100.0, !_fix_intercept);
- shark::KernelClassifier<shark::RealVector> model;
- shark::ZeroOneLoss<unsigned int> loss;
- shark::Data<unsigned int> output;
-
- int ii_train = 0;
- int ii_test = 0;
-
- for(int tt = 0; tt < _task_sizes_train.size(); ++tt)
- {
- trainer.stoppingCondition().maxSeconds = 0.1;
- trainer.train(model, training[tt]);
- if(_fix_intercept)
- {
- for(int cc = 0; cc < n_class - (n_class == 2); ++cc)
- {
- _coefs.push_back(std::vector<double>(_n_dim, 0.0));
- int sv = 0;
- for(auto el : model.decisionFunction().basis().elements())
- {
- daxpy_(_n_dim, model.parameterVector()(sv), &el(0), 1, _coefs.back().data(), 1);
- ++sv;
- }
- for(int dd = 0; dd < _n_dim; ++dd)
- {
- _coefs.back()[dd] = _feats[dd]->remap_w_svm(_coefs.back()[dd]);
- }
- }
- }
- else
- {
- for(int cc = 0; cc < n_class - (n_class == 2); ++cc)
- {
- _coefs.push_back(std::vector<double>(_n_dim + 1, 0.0));
- int sv = 0;
- for(auto el : model.decisionFunction().basis().elements())
- {
- daxpy_(_n_dim, model.parameterVector()(sv), &el(0), 1, _coefs.back().data(), 1);
- ++sv;
- }
- _coefs.back()[_n_dim] = model.parameterVector()(sv);
-
- for(int dd = 0; dd < _n_dim; ++dd)
- {
- _coefs.back()[dd] = _feats[dd]->remap_w_svm(_coefs.back()[dd]);
- _coefs.back()[_n_dim] -= _feats[dd]->remap_b_svm(_coefs.back()[dd]);
-
- }
- }
- }
- output = model(training[tt].inputs());
- for(auto element: output.elements())
- {
- _prop_train_est[ii_train] = static_cast<double>(element);
- ++ii_train;
- }
-
- std::vector<bool>train_misclassified(_n_samp_train);
- std::transform(_prop_train_est.begin(), _prop_train_est.end(), _prop_train.begin(), train_misclassified.begin(), [](double p1, double p2){return p1 != p2;});
- _train_n_svm_misclassified = std::accumulate(train_misclassified.begin(), train_misclassified.end(), 0);
-
- if(_task_sizes_test[tt] == 0)
- continue;
-
- output = model(test[tt].inputs());
- for(auto element: output.elements())
- {
- _prop_test_est[ii_test] = static_cast<double>(element);
- ++ii_test;
- }
- std::vector<bool>test_misclassified(_n_samp_train);
- std::transform(_prop_test_est.begin(), _prop_test_est.end(), _prop_train.begin(), test_misclassified.begin(), [](double p1, double p2){return p1 != p2;});
- _test_n_svm_misclassified = std::accumulate(test_misclassified.begin(), test_misclassified.end(), 0);
-
- }
+ std::vector<bool>train_misclassified(_n_samp_train);
+ std::transform(_prop_train_est.begin(), _prop_train_est.end(), _prop_train.begin(), train_misclassified.begin(), [](double p1, double p2){return p1 != p2;});
+ _train_n_svm_misclassified = std::accumulate(train_misclassified.begin(), train_misclassified.end(), 0);
+
+ std::vector<bool>test_misclassified(_n_samp_test);
+ std::transform(_prop_test_est.begin(), _prop_test_est.end(), _prop_test.begin(), test_misclassified.begin(), [](double p1, double p2){return p1 != p2;});
+ _test_n_svm_misclassified = std::accumulate(test_misclassified.begin(), test_misclassified.end(), 0);
+
+ // // Setup ML Shark data structures for SVM training
+ // std::vector<shark::LabeledData<shark::RealVector, unsigned int>> training(_task_sizes_train.size());
+ // std::vector<shark::LabeledData<shark::RealVector, unsigned int>> test(_task_sizes_test.size());
+ // std::vector<unsigned int> prop_to_copy(_prop_train.size());
+ // int n_els = 0;
+
+ // // Populate data structures for SVM and convex hull overlap
+ // for(int tt = 0; tt < _task_sizes_train.size(); ++tt)
+ // {
+ // shark::Data<shark::RealVector> data(_task_sizes_train[tt], shark::RealVector(_n_dim));
+ // shark::Data<unsigned int> labels(_task_sizes_train[tt], 0);
+
+ // for(int ii = 0; ii < _task_sizes_train[tt]; ++ii)
+ // prop_to_copy[ii] = static_cast<unsigned int>(_prop_train[ii + n_els]);
+
+ // int n_copied = 0;
+ // for(auto& batch : labels.batches())
+ // {
+ // std::copy_n(prop_to_copy.begin() + n_copied, shark::batchSize(batch), &batch(0));
+ // n_els += shark::batchSize(batch);
+ // n_copied += shark::batchSize(batch);
+ // }
+
+ // n_copied = 0;
+ // for(auto& batch : data.batches())
+ // {
+ // for(int ff = 0; ff < _n_dim; ++ff)
+ // dcopy_(shark::batchSize(batch), _feats[ff]->svm_value_ptr() + n_copied, 1, &batch(0, 0) + ff, _n_dim);
+
+ // n_copied += shark::batchSize(batch);
+ // }
+ // training[tt] = shark::LabeledData<shark::RealVector, unsigned int>(data, labels);
+ // }
+
+ // prop_to_copy.resize(_prop_test.size());
+ // n_els = 0;
+
+ // for(int tt = 0; tt < _task_sizes_test.size(); ++tt)
+ // {
+ // if(_task_sizes_test[tt] == 0)
+ // continue;
+
+ // shark::Data<shark::RealVector> data(_task_sizes_test[tt], shark::RealVector(_n_dim));
+ // shark::Data<unsigned int> labels(_task_sizes_test[tt], 0);
+
+ // for(int ii = 0; ii < _task_sizes_test[tt]; ++ii)
+ // prop_to_copy[ii] = static_cast<unsigned int>(_prop_test[ii + n_els]);
+
+ // int n_copied = 0;
+ // for(auto& batch : labels.batches())
+ // {
+ // std::copy_n(prop_to_copy.begin() + n_copied, shark::batchSize(batch), &batch(0));
+ // n_els += shark::batchSize(batch);
+ // n_copied += shark::batchSize(batch);
+ // }
+
+ // n_copied = 0;
+ // for(auto& batch : data.batches())
+ // {
+ // for(int ff = 0; ff < _n_dim; ++ff)
+ // dcopy_(shark::batchSize(batch), _feats[ff]->svm_test_value_ptr() + n_copied, 1, &batch(0, 0) + ff, _n_dim);
+
+ // n_copied += shark::batchSize(batch);
+ // }
+ // test[tt] = shark::LabeledData<shark::RealVector, unsigned int>(data, labels);
+ // }
+
+ // // Perform SVM training
+
+ // // Process SVM model
+ // std::shared_ptr<shark::AbstractKernelFunction<shark::RealVector>> _kernel_ptr = std::make_shared<shark::LinearKernel<shark::RealVector>>();
+ // shark::CSvmTrainer<shark::RealVector> trainer(_kernel_ptr.get(), 100.0, !_fix_intercept);
+ // shark::KernelClassifier<shark::RealVector> model;
+ // shark::ZeroOneLoss<unsigned int> loss;
+ // shark::Data<unsigned int> output;
+
+ // int ii_train = 0;
+ // int ii_test = 0;
+
+ // for(int tt = 0; tt < _task_sizes_train.size(); ++tt)
+ // {
+ // trainer.stoppingCondition().maxSeconds = 0.1;
+ // trainer.train(model, training[tt]);
+ // if(_fix_intercept)
+ // {
+ // for(int cc = 0; cc < n_class - (n_class == 2); ++cc)
+ // {
+ // _coefs.push_back(std::vector<double>(_n_dim, 0.0));
+ // int sv = 0;
+ // for(auto el : model.decisionFunction().basis().elements())
+ // {
+ // daxpy_(_n_dim, model.parameterVector()(sv), &el(0), 1, _coefs.back().data(), 1);
+ // ++sv;
+ // }
+ // for(int dd = 0; dd < _n_dim; ++dd)
+ // {
+ // _coefs.back()[dd] = _feats[dd]->remap_w_svm(_coefs.back()[dd]);
+ // }
+ // }
+ // }
+ // else
+ // {
+ // for(int cc = 0; cc < n_class - (n_class == 2); ++cc)
+ // {
+ // _coefs.push_back(std::vector<double>(_n_dim + 1, 0.0));
+ // int sv = 0;
+ // for(auto el : model.decisionFunction().basis().elements())
+ // {
+ // daxpy_(_n_dim, model.parameterVector()(sv), &el(0), 1, _coefs.back().data(), 1);
+ // ++sv;
+ // }
+ // _coefs.back()[_n_dim] = model.parameterVector()(sv);
+
+ // for(int dd = 0; dd < _n_dim; ++dd)
+ // {
+ // _coefs.back()[dd] = _feats[dd]->remap_w_svm(_coefs.back()[dd]);
+ // _coefs.back()[_n_dim] -= _feats[dd]->remap_b_svm(_coefs.back()[dd]);
+
+ // }
+ // }
+ // }
+ // output = model(training[tt].inputs());
+ // for(auto element: output.elements())
+ // {
+ // _prop_train_est[ii_train] = static_cast<double>(element);
+ // ++ii_train;
+ // }
+
+ // std::vector<bool>train_misclassified(_n_samp_train);
+ // std::transform(_prop_train_est.begin(), _prop_train_est.end(), _prop_train.begin(), train_misclassified.begin(), [](double p1, double p2){return p1 != p2;});
+ // _train_n_svm_misclassified = std::accumulate(train_misclassified.begin(), train_misclassified.end(), 0);
+
+ // if(_task_sizes_test[tt] == 0)
+ // continue;
+
+ // output = model(test[tt].inputs());
+ // for(auto element: output.elements())
+ // {
+ // _prop_test_est[ii_test] = static_cast<double>(element);
+ // ++ii_test;
+ // }
+ // std::vector<bool>test_misclassified(_n_samp_train);
+ // std::transform(_prop_test_est.begin(), _prop_test_est.end(), _prop_train.begin(), test_misclassified.begin(), [](double p1, double p2){return p1 != p2;});
+ // _test_n_svm_misclassified = std::accumulate(test_misclassified.begin(), test_misclassified.end(), 0);
+ // }
}
diff --git a/src/descriptor_identifier/Model/ModelClassifier.hpp b/src/descriptor_identifier/Model/ModelClassifier.hpp
index 50749f78..d2a416b8 100644
--- a/src/descriptor_identifier/Model/ModelClassifier.hpp
+++ b/src/descriptor_identifier/Model/ModelClassifier.hpp
@@ -18,12 +18,14 @@
#include<fstream>
#include<iostream>
-#include <shark/Algorithms/Trainers/CSvmTrainer.h>
-#include <shark/Models/Kernels/LinearKernel.h>
-#include <shark/Data/Dataset.h>
-#include <shark/ObjectiveFunctions/Loss/ZeroOneLoss.h>
+// #include <shark/Algorithms/Trainers/CSvmTrainer.h>
+// #include <shark/Models/Kernels/LinearKernel.h>
+// #include <shark/Data/Dataset.h>
+// #include <shark/ObjectiveFunctions/Loss/ZeroOneLoss.h>
#include <coin/ClpSimplex.hpp>
+
+#include <classification/SVMWrapper.hpp>
#include <descriptor_identifier/Model/Model.hpp>
#include <feature_creation/node/ModelNode.hpp>
#include <utils/string_utils.hpp>
diff --git a/src/external/libsvm/svm.cpp b/src/external/libsvm/svm.cpp
index fa44818b..e7724d6c 100644
--- a/src/external/libsvm/svm.cpp
+++ b/src/external/libsvm/svm.cpp
@@ -475,8 +475,8 @@ void Solver::reconstruct_gradient()
if(is_free(j))
nr_free++;
- if(2*nr_free < active_size)
- info("\nWARNING: using -h 0 may be faster\n");
+ // if(2*nr_free < active_size)
+ // info("\nWARNING: using -h 0 may be faster\n");
if (nr_free*l > 2*active_size*(l-active_size))
{
@@ -569,7 +569,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
{
counter = min(l,1000);
if(shrinking) do_shrinking();
- info(".");
+ // info(".");
}
int i,j;
@@ -579,7 +579,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
reconstruct_gradient();
// reset active set size and check
active_size = l;
- info("*");
+ // info("*");
if(select_working_set(i,j)!=0)
break;
else
@@ -735,7 +735,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
// reconstruct the whole gradient to calculate objective value
reconstruct_gradient();
active_size = l;
- info("*");
+ // info("*");
}
fprintf(stderr,"\nWARNING: reaching max number of iterations\n");
}
@@ -771,7 +771,7 @@ void Solver::Solve(int l, const QMatrix& Q, const double *p_, const schar *y_,
si->upper_bound_p = Cp;
si->upper_bound_n = Cn;
- info("\noptimization finished, #iter = %d\n",iter);
+ // info("\noptimization finished, #iter = %d\n",iter);
delete[] p;
delete[] y;
@@ -944,7 +944,7 @@ void Solver::do_shrinking()
unshrink = true;
reconstruct_gradient();
active_size = l;
- info("*");
+ // info("*");
}
for(i=0;i<active_size;i++)
@@ -1462,8 +1462,8 @@ static void solve_c_svc(
for(i=0;i<l;i++)
sum_alpha += alpha[i];
- if (Cp==Cn)
- info("nu = %f\n", sum_alpha/(Cp*prob->l));
+ // if (Cp==Cn)
+ // info("nu = %f\n", sum_alpha/(Cp*prob->l));
for(i=0;i<l;i++)
alpha[i] *= y[i];
@@ -1513,7 +1513,7 @@ static void solve_nu_svc(
alpha, 1.0, 1.0, param->eps, si, param->shrinking);
double r = si->r;
- info("C = %f\n",1/r);
+ // info("C = %f\n",1/r);
for(i=0;i<l;i++)
alpha[i] *= y[i]/r;
@@ -1590,7 +1590,7 @@ static void solve_epsilon_svr(
alpha[i] = alpha2[i] - alpha2[i+l];
sum_alpha += fabs(alpha[i]);
}
- info("nu = %f\n",sum_alpha/(param->C*l));
+ // info("nu = %f\n",sum_alpha/(param->C*l));
delete[] alpha2;
delete[] linear_term;
@@ -1625,7 +1625,7 @@ static void solve_nu_svr(
s.Solve(2*l, SVR_Q(*prob,*param), linear_term, y,
alpha2, C, C, param->eps, si, param->shrinking);
- info("epsilon = %f\n",-si->r);
+ // info("epsilon = %f\n",-si->r);
for(i=0;i<l;i++)
alpha[i] = alpha2[i] - alpha2[i+l];
@@ -1669,7 +1669,7 @@ static decision_function svm_train_one(
break;
}
- info("obj = %f, rho = %f\n",si.obj,si.rho);
+ // info("obj = %f, rho = %f\n",si.obj,si.rho);
// output SVs
@@ -1693,7 +1693,7 @@ static decision_function svm_train_one(
}
}
- info("nSV = %d, nBSV = %d\n",nSV,nBSV);
+ // info("nSV = %d, nBSV = %d\n",nSV,nBSV);
decision_function f;
f.alpha = alpha;
@@ -1805,13 +1805,13 @@ static void sigmoid_train(
if (stepsize < min_step)
{
- info("Line search fails in two-class probability estimates\n");
+ // info("Line search fails in two-class probability estimates\n");
break;
}
}
- if (iter>=max_iter)
- info("Reaching maximal iterations in two-class probability estimates\n");
+ // if (iter>=max_iter)
+ // info("Reaching maximal iterations in two-class probability estimates\n");
free(t);
}
@@ -1882,8 +1882,8 @@ static void multiclass_probability(int k, double **r, double *p)
}
}
}
- if (iter>=max_iter)
- info("Exceeds max_iter in multiclass_prob\n");
+ // if (iter>=max_iter)
+ // info("Exceeds max_iter in multiclass_prob\n");
for(t=0;t<k;t++) free(Q[t]);
free(Q);
free(Qp);
@@ -2003,7 +2003,7 @@ static double svm_svr_probability(
else
mae+=fabs(ymv[i]);
mae /= (prob->l-count);
- info("Prob. model for test data: target value = predicted value + z,\nz: Laplace distribution e^(-|z|/sigma)/(2sigma),sigma= %g\n",mae);
+ // info("Prob. model for test data: target value = predicted value + z,\nz: Laplace distribution e^(-|z|/sigma)/(2sigma),sigma= %g\n",mae);
free(ymv);
return mae;
}
@@ -2150,8 +2150,8 @@ svm_model *svm_train(const svm_problem *prob, const svm_parameter *param)
// group training data of the same class
svm_group_classes(prob,&nr_class,&label,&start,&count,perm);
- if(nr_class == 1)
- info("WARNING: training data in only one class. See README for details.\n");
+ // if(nr_class == 1)
+ // info("WARNING: training data in only one class. See README for details.\n");
svm_node **x = Malloc(svm_node *,l);
int i;
@@ -2270,7 +2270,7 @@ svm_model *svm_train(const svm_problem *prob, const svm_parameter *param)
nz_count[i] = nSV;
}
- info("Total nSV = %d\n",total_sv);
+ // info("Total nSV = %d\n",total_sv);
model->l = total_sv;
model->SV = Malloc(svm_node *,total_sv);
diff --git a/src/main.cpp b/src/main.cpp
index 0a209454..b6b12267 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -73,9 +73,7 @@ int main(int argc, char const *argv[])
}
}
}
- std::cout << "finialize mpi" << std::endl;
mpi_setup::finalize_mpi_env();
- std::cout << "out" << std::endl;
return 0;
}
diff --git a/src/python/descriptor_identifier/SISSOClassifier.cpp b/src/python/descriptor_identifier/SISSOClassifier.cpp
index 9f4019e7..f4ef0d30 100644
--- a/src/python/descriptor_identifier/SISSOClassifier.cpp
+++ b/src/python/descriptor_identifier/SISSOClassifier.cpp
@@ -22,14 +22,8 @@ SISSOClassifier::SISSOClassifier(
_lp_column_upper(_n_samp, COIN_DBL_MAX),
_lp_column_start(_n_samp, 0),
_lp_row((n_dim + 1) * _n_samp, 0),
- _training(_task_sizes_train.size()),
- _int_prop(_prop.size()),
- _int_prop_test(_prop_test.size()),
- _int_error(_prop.size()),
- _kernel_ptr(std::make_shared<shark::LinearKernel<shark::RealVector>>()),
- _trainer(_kernel_ptr.get(), 100.0, !_fix_intercept),
- _c(1.0),
- _width(1e-3),
+ _c(100.0),
+ _width(1.0e-5),
_n_class(1),
_lp_n_col(0),
_lp_n_row(0)
@@ -64,13 +58,11 @@ SISSOClassifier::SISSOClassifier(
_lp_column_upper(_n_samp, COIN_DBL_MAX),
_lp_column_start(_n_samp, 0),
_lp_row((n_dim + 1) * _n_samp, 0),
- _training(_task_sizes_train.size()),
- _int_prop(_prop.size()),
- _int_prop_test(_prop_test.size()),
- _int_error(_prop.size()),
- _kernel_ptr(std::make_shared<shark::LinearKernel<shark::RealVector>>()),
- _trainer(_kernel_ptr.get(), 100.0, !_fix_intercept),
- _c(1.0)
+ _c(100.0),
+ _width(1.0e-5),
+ _n_class(1),
+ _lp_n_col(0),
+ _lp_n_row(0)
{
std::vector<double> prop_vec = python_conv_utils::from_list<double>(prop);
std::vector<double> prop_test_vec = python_conv_utils::from_list<double>(prop_test);
--
GitLab