jazari is an educational machine learning library for C. Main purpose of the project is implementing some machine learning stuff from basic matrix operations to advanced models in a clean way.
What I cannot create, I do not understand. - Richard Feynman
p.s. It will be avoided to use 3rd-party libraries unless implementation is not related to machine learning concepts.
Creating a zero matrix with i x j size
Matrix* m = mcreate(i,j);
Converting 2d array to a matrix
double arr[2][3] = { {3,4,5},{8,1,5}};
Matrix* m = arr2matrix(3,2,arr);
Note that each sub-array corresponds to a vector in matrix. So you must reverse indices while creating matrix.
A[0][1] corresponds to A12 in mathematical notation.
Creating m x n diagonal matrix with number e
Matrix* m = mdiag(m,n,e);
Creating a n x n identity matrix
Matrix* m = ident(n);
Matrix* m = tranpose(n);
Matrix* m = add(m1,m2);
Matrix* m = dotp(m1,m2);
m1 and m2 are Matrix pointers in the examples above.
For now you can print matrix to console by print_matrix(m);
jazari presents a very simple usage to run linear regression model. Once you created your parameters and training data, running your model is as easy as 2 lines of code.
Firstly, initialize your model:
LRModel* lrm = init_linRM(params, training_data, targets, learning_rate, epoch_num);
In your Matrix* training_data, each column is an instance from dataset. Matrix* targets points to a vector in which each row is output of an instance.
To run your model, just write:
lrm->run();
Linear Regression Model uses Stochastic Gradient Descent (SGD) as its optimizer by default but you can replace it with your own custom optimizer.
Set optimizer after initialize your model:
lrm->run = your_own_optimizer;
Note that your_own_optimizer() must be a void function which only accepts one parameter in type of LRModel.