rmontanana/PyWrap
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Implement hyperparameters

Browse Source
This commit is contained in:
Ricardo Montañana Gómez
2023-11-08 10:35:38 +01:00
parent 1f46fc6c24
commit 331381930a
17 changed files with 913 additions and 325 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/CMakeLists.txt
View File

@@ -1,9 +1,8 @@
include_directories(${PyWrap_SOURCE_DIR}/lib/Files)
include_directories(${PyWrap_SOURCE_DIR}/lib/json/include)
include_directories(${Python3_INCLUDE_DIRS})
include_directories(${TORCH_INCLUDE_DIRS})
add_executable(main main.cc STree.cc SVC.cc RandomForest.cc PyClassifier.cc PyWrap.cc)
add_executable(example example.cpp PyWrap.cc)
target_link_libraries(main ${Python3_LIBRARIES} "${TORCH_LIBRARIES}" ${LIBTORCH_PYTHON} Boost::boost Boost::python Boost::numpy ArffFiles)
target_link_libraries(example ${Python3_LIBRARIES} "${TORCH_LIBRARIES}" ${LIBTORCH_PYTHON} Boost::boost Boost::python Boost::numpy ArffFiles)

13
src/Classifier.h Normal file
View File

@@ -0,0 +1,13 @@
#ifndef CLASSIFER_H
#define CLASSIFER_H
#include <nlohmann/json.hpp>
namespace pywrap {
class Classifier {
public:
Classifier() = default;
virtual ~Classifier() = default;
virtual void setHyperparameters(const nlohmann::json& hyperparameters) = 0;
};
} /* namespace pywrap */
#endif /* CLASSIFER_H */

22
src/PyClassifier.cc
View File

@@ -1,9 +1,10 @@
#include "PyClassifier.h"
#include <iostream>
namespace pywrap {
namespace bp = boost::python;
namespace np = boost::python::numpy;
PyClassifier::PyClassifier(const std::string& module, const std::string& className) : module(module), className(className)
PyClassifier::PyClassifier(const std::string& module, const std::string& className) : module(module), className(className), fitted(false)
{
pyWrap = PyWrap::GetInstance();
pyWrap->importClass(module, className);
@@ -36,10 +37,14 @@ namespace pywrap {
}
PyClassifier& PyClassifier::fit(torch::Tensor& X, torch::Tensor& y, const std::vector<std::string>& features, const std::string& className, std::map<std::string, std::vector<int>>& states)
{
if (!fitted && hyperparameters.size() > 0) {
std::cout << "Setting hyperparameters" << std::endl;
}
auto [Xn, yn] = tensors2numpy(X, y);
CPyObject Xp = bp::incref(bp::object(Xn).ptr());
CPyObject yp = bp::incref(bp::object(yn).ptr());
pyWrap->fit(module, this->className, Xp, yp);
fitted = true;
return *this;
}
torch::Tensor PyClassifier::predict(torch::Tensor& X)
@@ -69,4 +74,19 @@ namespace pywrap {
auto result = pyWrap->score(module, className, Xp, yp);
return result;
}
void PyClassifier::setHyperparameters(const nlohmann::json& hyperparameters)
{
// Check if hyperparameters are valid, default is no hyperparameters
const std::vector<std::string> validKeys = { };
checkHyperparameters(validKeys, hyperparameters);
this->hyperparameters = hyperparameters;
}
void PyClassifier::checkHyperparameters(const std::vector<std::string>& validKeys, const nlohmann::json& hyperparameters)
{
for (const auto& item : hyperparameters.items()) {
if (find(validKeys.begin(), validKeys.end(), item.key()) == validKeys.end()) {
throw std::invalid_argument("Hyperparameter " + item.key() + " is not valid");
}
}
}
} /* namespace pywrap */

10
src/PyClassifier.h
View File

@@ -2,15 +2,17 @@
#define PYCLASSIFER_H
#include "boost/python/detail/wrap_python.hpp"
#include <boost/python/numpy.hpp>
#include <nlohmann/json.hpp>
#include <string>
#include <map>
#include <vector>
#include <utility>
#include <torch/torch.h>
#include "PyWrap.h"
#include "Classifier.h"
namespace pywrap {
class PyClassifier {
class PyClassifier : public Classifier {
public:
PyClassifier(const std::string& module, const std::string& className);
virtual ~PyClassifier();
@@ -19,11 +21,15 @@ namespace pywrap {
double score(torch::Tensor& X, torch::Tensor& y);
std::string version();
std::string callMethodString(const std::string& method);
void setHyperparameters(const nlohmann::json& hyperparameters) override;
protected:
void checkHyperparameters(const std::vector<std::string>& validKeys, const nlohmann::json& hyperparameters);
nlohmann::json hyperparameters;
private:
PyWrap* pyWrap;
std::string module;
std::string className;
bool fitted;
};
} /* namespace pywrap */
#endif /* PYCLASSIFER_H */

7
src/STree.cc
View File

@@ -5,4 +5,11 @@ namespace pywrap {
{
return callMethodString("graph");
}
void STree::setHyperparameters(const nlohmann::json& hyperparameters)
{
// Check if hyperparameters are valid
const std::vector<std::string> validKeys = { "C", "n_jobs", "kernel", "max_iter", "max_depth", "random_state", "multiclass_strategy" };
checkHyperparameters(validKeys, hyperparameters);
this->hyperparameters = hyperparameters;
}
} /* namespace pywrap */

2
src/STree.h
View File

@@ -1,5 +1,6 @@
#ifndef STREE_H
#define STREE_H
#include "nlohmann/json.hpp"
#include "PyClassifier.h"
namespace pywrap {
@@ -8,6 +9,7 @@ namespace pywrap {
STree() : PyClassifier("stree", "Stree") {};
~STree() = default;
std::string graph();
void setHyperparameters(const nlohmann::json& hyperparameters) override;
};
} /* namespace pywrap */
#endif /* STREE_H */

257
src/example.cpp
View File

@@ -1,257 +0,0 @@
#include "boost/python/detail/wrap_python.hpp"
#include <boost/python/numpy.hpp>
#include <torch/torch.h>
#include <torch/csrc/utils/tensor_numpy.h>
#include <string>
#include <iostream>
#include "ArffFiles.h"
#include "PyHelper.hpp"
#include "PyWrap.h"
void errorAbort(const std::string& message)
{
std::cerr << message << std::endl;
PyErr_Print();
exit(1);
}
void print_array(pywrap::np::ndarray& array)
{
std::cout << "Array: " << std::endl;
std::cout << pywrap::p::extract<char const*>(pywrap::p::str(array)) << std::endl;
}
// np::ndarray to_numpy_matrix(torch::Tensor& input_data, np::dtype numpy_dtype)
// {
// p::tuple shape = p::make_tuple(input_data.size(0), input_data.size(1));
// auto tensor_dtype = input_data.dtype();
// p::tuple stride = p::make_tuple(sizeof(tensor_dtype) * input_data.size(1), sizeof(tensor_dtype));
// auto dito = input_data.transpose(1, 0);
// np::ndarray result = np::from_data(dito.data_ptr(), numpy_dtype, shape, stride, p::object());
// return result;
// }
// np::ndarray to_numpy_vector(torch::Tensor& input_data, np::dtype numpy_dtype)
// {
// p::tuple shape = p::make_tuple(input_data.size(0));
// auto tensor_dtype = input_data.dtype();
// p::tuple stride = p::make_tuple(sizeof(tensor_dtype), sizeof(tensor_dtype));
// np::ndarray result = np::from_data(input_data.data_ptr(), numpy_dtype, shape, stride, p::object());
// return result;
// }
class Paths {
public:
static string datasets()
{
return "../discretizbench/datasets/";
}
};
tuple<torch::Tensor, torch::Tensor, vector<string>, string, map<string, vector<int>>> loadDataset(const string& name, bool class_last)
{
auto handler = ArffFiles();
handler.load(Paths::datasets() + static_cast<string>(name) + ".arff", class_last);
// Get Dataset X, y
vector<vector<float>> X = handler.getX();
vector<int> y = handler.getY();
// Get className & Features
auto className = handler.getClassName();
vector<string> features;
auto attributes = handler.getAttributes();
transform(attributes.begin(), attributes.end(), back_inserter(features), [](const auto& pair) { return pair.first; });
torch::Tensor Xd;
auto states = map<string, vector<int>>();
Xd = torch::zeros({ static_cast<int>(X.size()), static_cast<int>(X[0].size()) }, torch::kFloat32);
for (int i = 0; i < features.size(); ++i) {
Xd.index_put_({ i, "..." }, torch::tensor(X[i], torch::kFloat32));
}
return { Xd, torch::tensor(y, torch::kInt32), features, className, states };
}
using namespace pywrap;
np::ndarray tensor2numpy(torch::Tensor& X)
{
int m = X.size(0);
int n = X.size(1);
auto Xn = np::from_data(X.data_ptr(), np::dtype::get_builtin<float>(), p::make_tuple(m, n), p::make_tuple(sizeof(X.dtype()) * 2 * n, sizeof(X.dtype()) * 2), p::object());
Xn = Xn.transpose();
return Xn;
}
pair<np::ndarray, np::ndarray> tensors2numpy(torch::Tensor& X, torch::Tensor& y)
{
int n = X.size(1);
auto yn = np::from_data(y.data_ptr(), np::dtype::get_builtin<int32_t>(), p::make_tuple(n), p::make_tuple(sizeof(y.dtype()) * 2), p::object());
return { tensor2numpy(X), yn };
}
pair<np::ndarray, np::ndarray> getData(const string& dataset)
{
auto [X, y, featuresx, classNamex, statesx] = loadDataset(dataset, true);
auto [Xn, yn] = tensors2numpy(X, y);
auto Xn_shapes = Xn.get_shape();
auto yn_shapes = yn.get_shape();
cout << "Xn_shapes: " << Xn_shapes[0] << ", " << Xn_shapes[1] << endl;
cout << "yn_shapes: " << yn_shapes[0] << endl;
cout << "X shapes: " << X.sizes() << endl;
cout << "y shapes: " << y.sizes() << endl;
assert(Xn_shapes[0] == X.sizes()[0]);
assert(Xn_shapes[1] == X.sizes()[1]);
assert(yn_shapes[0] == y.sizes()[0]);
return { Xn, yn };
}
int main(int argc, char** argv)
{
cout << "* Begin." << endl;
{
PyWrap* wrapper = PyWrap::GetInstance();
string dataset = "iris";
// Convert Tensor to numpy array
// auto [Xn, yn] = tensors2numpy(X, y);
// cout << "Numpy array data: " << endl;
// print_array(Xn);
// cout << "Numpy array labels: " << endl;
// print_array(yn);
// Import module
string moduleName = "stree";
string className = "Stree";
// Import
{
cout << "--Import Phase--" << endl;
wrapper->importClass(moduleName, className);
cout << "--Import Phase end--" << endl;
}
// Version
{
cout << "--Version Phase--" << endl;
auto version = wrapper->version(moduleName, className);
cout << "Version: " << version << endl;
cout << "--Version Phase end--" << endl;
}
// Fit
{
cout << "--Fit Phase--" << endl;
auto [Xn, yn] = getData(dataset);
auto Xn_shapes = Xn.get_shape();
auto yn_shapes = yn.get_shape();
CPyObject Xp = boost::python::incref(boost::python::object(Xn).ptr());
CPyObject yp = boost::python::incref(boost::python::object(yn).ptr());
//print_array(yn);
// Call fit
cout << "Calling fit" << endl;
wrapper->fit(moduleName, className, Xp, yp);
cout << "--Fit Phase end--" << endl;
}
// Score
{
cout << "--Score Phase--" << endl;
auto [Xn, yn] = getData(dataset);
auto Xn_shapes = Xn.get_shape();
auto yn_shapes = yn.get_shape();
CPyObject Xp = boost::python::incref(boost::python::object(Xn).ptr());
CPyObject yp = boost::python::incref(boost::python::object(yn).ptr());
//print_array(yn);
// Call score
cout << "Calling score" << endl;
auto result = wrapper->score(moduleName, className, Xp, yp);
cout << "Score: " << result << endl;
cout << "--Score Phase end--" << endl;
}
// Call score
// {
// np::initialize();
// cout << "--Score Phase--" << endl;
// auto [X, y, featuresx, classNamex, statesx] = loadDataset(dataset, true);
// auto [Xn, yn] = tensors2numpy(X, y);
// auto Xn_shapes = Xn.get_shape();
// auto yn_shapes = yn.get_shape();
// cout << "Xn_shapes: " << Xn_shapes[0] << ", " << Xn_shapes[1] << endl;
// cout << "yn_shapes: " << yn_shapes[0] << endl;
// cout << "X shapes: " << X.sizes() << endl;
// cout << "y shapes: " << y.sizes() << endl;
// assert(Xn_shapes[0] == X.sizes()[0]);
// assert(Xn_shapes[1] == X.sizes()[1]);
// assert(yn_shapes[0] == y.sizes()[0]);
// CPyObject Xp = Xn.ptr();
// CPyObject yp = yn.ptr();
// print_array(yn);
// cout << "Calling score" << endl;
// auto instance = wrapper->getClass(moduleName, className);
// CPyObject result;
// if (!(result = PyObject_CallMethod(instance, "score", "OO", Xp.getObject(), yp.getObject())))
// errorAbort("Couldn't call method score");
// auto score = PyFloat_AsDouble(result);
// //auto score = wrapper->score(moduleName, className, Xp, yp);
// cout << "Score: " << score << endl;
// cout << "--Score Phase end--" << endl;
// }
// Clean module
{
cout << "--Clean Phase--" << endl;
wrapper->clean(moduleName, className);
cout << "--Clean Phase end--" << endl;
}
}
cout << "* End." << endl;
}
// int main(int argc, char** argv)
// {
// auto [data_tensor, y_label, featuresx, classNamex, statesx] = loadDataset("iris", true);
// // CPyInstance pInstance;
// // auto wrapper = PyWrap();
// PyWrap* wrapper = PyWrap::GetInstance();
// // PyWrap* wrapper = PyWrap::GetInstance();
// int m = data_tensor.size(0);
// int n = data_tensor.size(1);
// auto data_numpy = np::from_data(data_tensor.data_ptr(), np::dtype::get_builtin<float>(), p::make_tuple(m, n), p::make_tuple(sizeof(data_tensor.dtype()) * 2 * n, sizeof(data_tensor.dtype()) * 2), p::object());
// data_numpy = data_numpy.transpose();
// auto y_numpy = np::from_data(y_label.data_ptr(), np::dtype::get_builtin<int32_t>(), p::make_tuple(n), p::make_tuple(sizeof(y_label.dtype()) * 2), p::object());
// cout << "Numpy array data: " << endl;
// print_array(data_numpy);
// cout << "Numpy array labels: " << endl;
// print_array(y_numpy);
// cout << "primero" << endl;
// CPyObject p = data_numpy.ptr();
// CPyObject yp = y_numpy.ptr();
// string moduleName = "sklearn.svm";
// string className = "SVC";
// string method = "_repr_html_";
// // CPyObject module = PyImport_ImportModule(moduleName.c_str());
// // if (PyErr_Occurred()) {
// // errorAbort("Could't import module " + moduleName);
// // }
// // CPyObject classObject = PyObject_GetAttrString(module, className.c_str());
// // if (PyErr_Occurred()) {
// // errorAbort("Couldn't find class " + className);
// // }
// // CPyObject instance = PyObject_CallObject(classObject, NULL);
// // if (PyErr_Occurred()) {
// // errorAbort("Couldn't create instance of class " + className);
// // }
// // wrapper.moduleClassMap.insert({ { moduleName, className }, { module, classObject, instance } });
// wrapper->importClass(moduleName, className);
// PyObject* instance = wrapper->getClass(moduleName, className);
// CPyObject result;
// if (!(result = PyObject_CallMethod(instance, method.c_str(), NULL)))
// errorAbort("Couldn't call method " + method);
// std::string value = PyUnicode_AsUTF8(result);
// cout << "Version: " << value << endl;
// cout << "Calling fit" << endl;
// p.AddRef();
// yp.AddRef();
// method = "fit";
// wrapper->fit(moduleName, className, p, yp);
// // PyObject* instance2 = wrapper->getClass(moduleName, className);
// // if (!(result = PyObject_CallMethodObjArgs(instance2, PyUnicode_FromString(method.c_str()), p.getObject(), yp.getObject(), NULL)))
// // errorAbort("Couldn't call method fit");
// // method = "fit";
// // if (!(result = PyObject_CallMethodObjArgs(instance, PyUnicode_FromString(method.c_str()), p.getObject(), yp.getObject(), NULL)))
// // errorAbort("Couldn't call method fit");
// cout << "Calling score" << endl;
// // method = "score";
// // if (!(result = PyObject_CallMethodObjArgs(instance, PyUnicode_FromString(method.c_str()), p.getObject(), yp.getObject(), NULL)))
// // errorAbort("Couldn't call method score");
// // float score = PyFloat_AsDouble(result);
// auto score = wrapper->score(moduleName, className, p, yp);
// cout << "Score: " << score << endl;
// wrapper->clean(moduleName, className);
// return 0;
// }

2
src/main.cc
View File

@@ -52,6 +52,8 @@ int main(int argc, char* argv[])
cout << "X: " << X.sizes() << endl;
cout << "y: " << y.sizes() << endl;
auto clf = pywrap::STree();
auto hyperparameters = nlohmann::json({ "max_depth": 3, "C" : 0.7 });
clf.setHyperparameters(hyperparameters);
cout << "STree Version: " << clf.version() << endl;
auto svc = pywrap::SVC();
svc.fit(X, y, features, className, states);