Machine Learning Models

Some ML models for classification and regression implemented from scratch in C.

Plots

Classification

PLA	POCKET	ADALINE

Regression

Simple linear regression (Least squares with gradient decent)	Evolution of loss

Getting started

Dependencies (Optionnal)

• gnuplot (optionnal) This is just for the plot.sh script, you can compile and use the library without this
• rainy-csv (optionnal) This is my library, I used it just for the tests, so you can skip it too, it is used as git submodule so it will be automatically downloaded when you clone with --recurse-submodules

Cloning

If you are planning to run the tests, please clone with

$ git clone --recurse-submodules https://github.com/hazyuun/machine-learning-models

to automatically get rainy-csv.
If you won't run the tests, then clone as usual :

$ git clone https://github.com/hazyuun/machine-learning-models

Compiling and using

Compiling

Just use the Makefile

Compile everthing

$ make USE_RAINY_CSV=1 all

Output will be in lib for the library, and in bin for the tests, note that USE_RAINY_CSV macro needs to be defined to compile the tests.

Only compile the library

if you want to compile the library without rainy-csv support, run :

$ make lib

otherwise, run

$ make USE_RAINY_CSV=1 lib

Output will be in lib directory

Compile the test (supposing you already compiled the library

$ make test

Output will be in bin directory

Usage example

#include <helpers/model.h>
#include <helpers/csv2dataset.h>
#include <utils/export.h>

int main(int argc, char **argv) {
  (void)argc;
  (void)argv;

   /* Load data */
  labeled_dataset_t *data;
  data = csv_to_labeled_dataset("./test-data/noisy01.csv");

  /* Make and train the model */
  single_layer_model_t *model;
  model = make_single_layer_model(LOG_REG_MODEL, 2, &sigmoid);
  history_t *history = model->train(model, data);

  /* Generate the output files for plotting */
  gp_export_labeled_dataset(data, "data.in");
  gp_export_single_layer_model(model, history);
}

Running the tests

First of all, after compiling, run the test of your choice, eg.

$ ./bin/rg_simple_linreg_test

Some output files will be generated, those files are used by the gnuplot scripts to show the result

Run the plot.sh bash script to see the plots

• Reminder : you will need gnuplot for that

For the classification tests run :

$ ./plot.sh

For the linear regression ones run :

$ ./plot.sh --lin-reg

Notes about the tests

• Test files prefixed with cl_ are for classification, and ones prefixed with rg_ are for regression.