CPP-ML

Implementations of machine learning algorithms.

Implemented algorithms

• K-Nearest Neighbour (KNN)
• Linear Regression
• Logistic Regression
• Decision Tree

Dependencies

• Eigen (v3.3.9) See here
• matplotplusplus (v1.2.0) See here

Usage

Clone project

git clone https://github.com/jasperingo/cpp-ml.git

Set up dependecies for CMake as describe in their respective documentations

Build project with CMake

Run executable

Arguments

--dataset

Path to data set file

--dataset /home/student-test-score.csv

--algorithm

ML algorithm to use. Values can be any of:

• knn
• linear-regression
• logistic-regression
• decision-tree

--algorithm knn

--label-column

Index of the column in the dataset that contains the values for sample labels

Indexes start from 0

--label-column 1

--label-column-digit

Indicates if the value of the labels are digit or not. Values can be:

• true / 1
• false / 0

--label-column-digit true

--feature-columns

Comma separated indexes of the columns in the dataset that contains the values for sample features

Indexes start from 0

--feature-columns 1,2,4

--feature-columns-digit

Comma separated values that indicates if the value of the features are digit or not. Values can be:

• true / 1
• false / 0

--feature-columns-digit true,false,true

--number-of-k

Optional argument

Used to specify a custom number of K neighbours for KNN algorithm. Default value is 3

--number-of-k 5

--learning-rate

Optional argument

Used to specify a custom learning rate for algorithms that support it. Default value is 0.001

--learning-rate 0.01

--max-iterations

Optional argument

Used to specify a custom maximum number of iterations for algorithms that support it. Default value is 1000

--max-iterations 2000

--max-depth

Optional argument

Used to specify a custom maximum tree depth for algorithms that support it. Default value is 100

--max-depth 200

--min-sample-split

Optional argument

Used to specify a custom minimum number of sample to be present in a tree node for splitting of the node to be allowed depth for algorithms that support it. Default value is 2

--min-sample-split 5

--number-of-features

Optional argument

Used to specify a custom number of features to selected at random from the provided features at specific point during training for algorithms that support it. Default value is number of features - 1

--number-of-features 3

--number-of-trees

Optional argument

Used to specify a custom number of trees for Random forest algorithm. Default value is 5

--number-of-trees 10