Machine Learning

Machine learning is about optimization. In learning via optimization, we define an error function (loss function) on the model and try to optimize it, i.e., try to find the optimal parameters of the model by optimization techniques borrowed from optimization literature.

Machine learning is about algorithms. In decision/regression trees, we need to write a recursive-learning algorithm to classify the data arriving at each decision node, and we also need to write a recursive code to generate the decision tree structure.

Machine learning is about statistics. We usually assume Gaussian noise on the data, we assume multivariate normal distribution on class covariance matrices in quadratic discriminant analysis, and we use cross-validation or bootstrapping to generate multiple training sets.

Machine learning is about models. In decision trees, the data structure is a binary/L-ary tree depending on the type of the features and the decision function used.

Machine learning is about performance metrics. In classification, we use accuracy if we want to get a crude estimate on the performance of the classifier. If we need more details on pairwise classes, confusion matrix comes in handy. If the dataset has two classes, more metrics follow: precision, recall, true positive rate, false positive rate, F-measure, etc.

What is machine learning about? Like defining an elephant, we need the combination of all of these topics to define machine learning: we start machine learning by assuming a certain model on the data, use algorithmic and/or optimization and/or statistical techniques to learn the parameters of that model (which is sometimes defined as curve fitting), and use performance metrics to evaluate our model/algorithm/classifier.

Algorithms

Nearest Neighbor

The most commonly used representative of nonparametric algorithms is the nearest neighbor. The assumption of nearest neighbor is simple, the world does not change much, i.e., similar things perform similarly. Therefore, we only need to store the dataset itself and make the decision on the test instance based on the similarity of it to the instances in the dataset. In other words, the class label of an instance is strongly influenced by its nearby instances.

Parametric Classification

If the class distributions are assumed to follow Gaussian density, we obtain our first parametric classifier, namely, quadratic discriminant. The number of parameters, i.e., the model complexity is Kd + Kd (d + 1) / 2, the first part is for class means and the second part for class covariance matrices.

We can assume a single shared covariance matrix for all classes. In this case, simplifying the function reduces to our second classifier, namely, linear discriminant. The model complexity is Kd + d (d + 1) / 2, where the first part is for class means and the second part for shared covariance matrix.

When we assume all off-diagonals of the shared covariance matrix are zero; we get the naive Bayes classifier. The model complexity is Kd + d, where the first part is for class means and the second part for the diagonal of the shared covariance matrix.

We further reduce by taking priors equal and a single covariance value s. In this case, we get the nearest mean classifier and the model complexity is only Kd + 1.

Decision Trees

Decision trees have a tree-based structure where each non-leaf node m implements a decision function, f_m(x), and each leaf node corresponds to a class decision. Second, they are one of most interpretable learning algorithms available. When written as a set of IF-THEN rules, the decision tree can be transformed into a human-readable format, which then can be modified and/or validated by human experts in their corresponding domains.

Kernel Machines

Kernel machines, in other words, support vector machines, are maximum margin methods, where the model is written as a weighted sum of support vectors. Kernel machines are discriminative methods, i.e., they are only interested in the instances across the class boundaries in classification, or instances across the regressor in regression. For obtaining the optimal separating hyperplane, kernel machines try to maximize separability, or margin, and write the problem as a quadratic optimization problem, whose solution gives us support vectors.

Neural Networks

Artificial neural networks (ANN) take their inspiration from the brain. The brain consists of billions of neurons and these neurons are interconnected and work in parallel, which makes the brain a powerful computing machine. Each neuron is connected through synapses to thousands of neurons and the firing of a neuron depends on those synapses.

There are three types of neurons (units) in ANN. Each unit except the input unit takes an input and calculates an output. Input units represent a single input feature x_i or the bias = +1. Hidden units calculate an intermediate output from its inputs. They first combine their inputs linearly and then use nonlinear activation functions to map that linear combination to a nonlinear space. Output units calculate the output of the ANN.

Video Lectures

For Developers

You can also see Java, Python, Cython, C, Swift, Js, or C# repository.

Requirements

• C++ Compiler
• Git

CPP

To check if you have compatible C++ Compiler installed,

• Open CLion IDE
• Preferences >Build,Execution,Deployment > Toolchain

Git

Install the latest version of Git.

Download Code

In order to work on code, create a fork from GitHub page. Use Git for cloning the code to your local or below line for Ubuntu:

git clone <your-fork-git-link>

A directory called Classification-CPP will be created. Or you can use below link for exploring the code:

git clone https://github.com/starlangsoftware/Classification-CPP.git

Open project with CLion IDE

To import projects from Git with version control:

• Open CLion IDE , select Get From Version Control.

• In the Import window, click URL tab and paste github URL.

• Click open as Project.

Result: The imported project is listed in the Project Explorer view and files are loaded.

Compile

From IDE

After being done with the downloading and opening project, select Build Project option from Build menu. After compilation process, user can run TestClassification.cpp.

Detailed Description

• Classification Algorithms
• Sampling Strategies
• Feature Selection
• Statistical Tests

Classification Algorithms

Algoritmaları eğitmek için

void train(InstanceList trainSet, Parameter parameters)

Eğitilen modeli bir veri örneği üstünde sınamak için

String predict(Instance instance)

Karar ağacı algoritması C45 sınıfında

Bagging algoritması Bagging sınıfında

Derin öğrenme algoritması DeepNetwork sınıfında

KMeans algoritması KMeans sınıfında

Doğrusal ve doğrusal olmayan çok katmanlı perceptron LinearPerceptron ve MultiLayerPerceptron sınıflarında

Naive Bayes algoritması NaiveBayes sınıfında

K en yakın komşu algoritması Knn sınıfında

Doğrusal kesme analizi algoritması Lda sınıfında

İkinci derece kesme analizi algoritması Qda sınıfında

Destek vektör makineleri algoritması Svm sınıfında

RandomForest ağaç tabanlı ensemble algoritması RandomForest sınıfında

Basit dummy ve rasgele sınıflandırıcı gibi temel iki sınıflandırıcı Dummy ve RandomClassifier sınıflarında

Sampling Strategies

K katlı çapraz geçerleme deneyi yapmak için KFoldRun, KFoldRunSeparateTest, StratifiedKFoldRun, StratifiedKFoldRunSeparateTest

M tane K katlı çapraz geçerleme deneyi yapmak için MxKFoldRun, MxKFoldRunSeparateTest, StratifiedMxKFoldRun, StratifiedMxKFoldRunSeparateTest

Bootstrap tipi deney yapmak için BootstrapRun

Feature Selection

Pca tabanlı boyut azaltma için Pca sınıfı

Discrete değişkenleri Continuous değişkenlere çevirmek için DiscreteToContinuous sınıfı

Discrete değişkenleri binary değişkenlere değiştirmek için LaryToBinary sınıfı

Statistical Tests

İstatistiksel testler için Combined5x2F, Combined5x2t, Paired5x2t, Pairedt, Sign sınıfları

For Contibutors

Conan Setup

1. First install conan.

pip install conan

Instructions are given in the following page:

https://docs.conan.io/2/installation.html

2. Add conan remote 'ozyegin' with IP: 104.247.163.162 with the following command:

conan remote add ozyegin http://104.247.163.162:8081/artifactory/api/conan/conan-local --insert

3. Use the comman conan list to check for installed packages. Probably there are no installed packages.

conan list

conanfile.py file

1. Put the correct dependencies in the requires part

    requires = ["math/1.0.0", "classification/1.0.0"]

2. Default settings are:

    settings = "os", "compiler", "build_type", "arch"
    options = {"shared": [True, False], "fPIC": [True, False]}
    default_options = {"shared": True, "fPIC": True}
    exports_sources = "src/*", "Test/*"

    def layout(self):
        cmake_layout(self, src_folder="src")

    def generate(self):
        tc = CMakeToolchain(self)
        tc.generate()
        deps = CMakeDeps(self)
        deps.generate()

    def build(self):
        cmake = CMake(self)
        cmake.configure()
        cmake.build()

    def package(self):
        copy(conanfile=self, keep_path=False, src=join(self.source_folder), dst=join(self.package_folder, "include"), pattern="*.h")
        copy(conanfile=self, keep_path=False, src=self.build_folder, dst=join(self.package_folder, "lib"), pattern="*.a")
        copy(conanfile=self, keep_path=False, src=self.build_folder, dst=join(self.package_folder, "lib"), pattern="*.so")
        copy(conanfile=self, keep_path=False, src=self.build_folder, dst=join(self.package_folder, "lib"), pattern="*.dylib")
        copy(conanfile=self, keep_path=False, src=self.build_folder, dst=join(self.package_folder, "bin"), pattern="*.dll")

    def package_info(self):
        self.cpp_info.libs = ["ComputationalGraph"]

CMakeLists.txt file

1. Set the C++ standard with compiler flags.

	set(CMAKE_CXX_STANDARD 20)
	set(CMAKE_CXX_FLAGS "-O3")

2. Dependent packages should be given with find_package.

	find_package(util_c REQUIRED)
	find_package(data_structure_c REQUIRED)

3. For library part, use add_library and target_link_libraries commands. Use m library for math linker in Linux.

	add_library(Math src/Distribution.cpp src/Distribution.h src/DiscreteDistribution.cpp src/DiscreteDistribution.h src/Vector.cpp src/Vector.h src/Eigenvector.cpp src/Eigenvector.h src/Matrix.cpp src/Matrix.h src/Tensor.cpp src/Tensor.h)
	target_link_libraries(Math util_c::util_c data_structure_c::data_structure_c m)

4. For executable tests, use add_executable and target_link_libraries commands. Use m library for math linker in Linux.

	add_executable(DiscreteDistributionTest src/Distribution.cpp src/Distribution.h src/DiscreteDistribution.cpp src/DiscreteDistribution.h src/Vector.cpp src/Vector.h src/Eigenvector.cpp src/Eigenvector.h src/Matrix.cpp src/Matrix.h src/Tensor.cpp src/Tensor.h Test/DiscreteDistributionTest.cpp)
	target_link_libraries(DiscreteDistributionTest util_c::util_c data_structure_c::data_structure_c m)

Data files

1. Add data files to the cmake-build-debug folder.

C++ files

1. If needed, comparator operators == and < should be implemented for map and set data structures.

    bool operator==(const Word &anotherWord) const{
        return (name == anotherWord.name);
    }
    bool operator<(const Word &anotherWord) const{
        return (name < anotherWord.name);
    }

2. Do not forget to comment each function.

	/**
 	* A constructor of Word class which gets a String name as an input and assigns to the name variable.
	*
	* @param _name String input.
 	*/
	Word::Word(const string &_name) {

3. Function names should follow caml case.

	int Word::charCount() const

4. Write getter and setter methods.

	string Word::getName() const
	void Word::setName(const string &_name)

5. Use catch.hpp for testing purposes. Add

#define CATCH_CONFIG_MAIN  // This tells Catch to provide a main() - only do this in one cpp file

line in only one of the test files. Add

#include "catch.hpp"

line in all test files. Example test file is given below:

TEST_CASE("DictionaryTest") {
    TxtDictionary lowerCaseDictionary = TxtDictionary("lowercase.txt", "turkish_misspellings.txt");
    TxtDictionary mixedCaseDictionary = TxtDictionary("mixedcase.txt", "turkish_misspellings.txt");
    TxtDictionary dictionary = TxtDictionary();
    SECTION("testSize"){
        REQUIRE(29 == lowerCaseDictionary.size());
        REQUIRE(58 == mixedCaseDictionary.size());
        REQUIRE(62113 == dictionary.size());
    }
    SECTION("testGetWord"){
        for (int i = 0; i < dictionary.size(); i++){
            REQUIRE_FALSE(nullptr == dictionary.getWord(i));
        }
    }
    SECTION("testLongestWordSize"){
        REQUIRE(1 == lowerCaseDictionary.longestWordSize());
        REQUIRE(1 == mixedCaseDictionary.longestWordSize());
        REQUIRE(21 == dictionary.longestWordSize());
    }

6. Enumerated types should be declared with enum class.

	enum class Pos {
		ADJECTIVE,
		NOUN,
		VERB,
		ADVERB,

7. Every header file should start with

	#ifndef MATH_DISTRIBUTION_H
	#define MATH_DISTRIBUTION_H

and end with

	#endif //MATH_DISTRIBUTION_H

8. Do not forget to use const expression for parameters, if they will not be changed in the function.

	void Word::setName(const string &_name);

9. Do not forget to use const expression for methods, which do not modify any class attribute. Also use [[dodiscard]]

	[[nodiscard]] bool isPunctuation() const;

10. Use xmlparser package for parsing xml files.

    auto* doc = new XmlDocument("test.xml");
    doc->parse();
    XmlElement* root = doc->getFirstChild();
    XmlElement* firstChild = root->getFirstChild();

11. Data structures: Use map for hash map, unordered_map for linked hash map, vector for array list, unordered_set for hash set