madrury / linalg

A linear algebra library in C. For fun.

linalg: Linear Algebra and Regression in C

linalg is a library for linear algebra and regression implemented in C. The code is optimized for readability and clarity instead of raw efficiency (though it tries not to ignore issues of efficiency completely).

Linear Algebra

linalg contains two datatypes in its core linear algebra engine, vector and matrix:

• vector is a (one dimensional) vector of real numbers (C doubles). The underlying data is stored as a C array, and so occupies contiguous memory locations in the computer's memory.
• matrix is a (two dimensional) matrix of real numbers (C doubles). The underlying data is stored in row-major order, so each row occupies contiguous memory locations in the computer's memory.

To complement these data types, linalg contains many functions for performing linear algebraic operations. For example

• matrix_vector_multiply computes the product vector of a matrix and vector.
• matrix_multiply computes the product matrix of two matrices.
• matrix_multiply_MtN computes the product of the transpose of one matrix with another.

Linear equations can be solved using linsolve_qr, which adopts a strategy of computing the QR matrix factorization of the left hand side. To access the underlying matrix factorization, use qr_decomp.

Regression

linalg also includes functions for regression. Use linreg_fit to fit a linear regression given a design matrix X and a response vector y.

Tests

The routines in linalg are extensively unit tested, which also serve as simple examples of use.

linalg's Issues

Add test names to testing suite.

Make a test a struct {test_function, test_name}.

Remove TODOs for argument consistency checks.

License

Could you clarify the license?

Check Upper Triangular

in linsolve_upper_triangular

matrix_from_array and vector_from_array have conflicting apis.

Argument order differers.

Implement cache blocking for multiplication.

Move tests from main into a test file which automatically checks success.

Implement equality comparison for vector and matrix.

Need a tolerance bound for this method.

`eigenfree` does not free matrix of eigenvectors.

Implement cache blocking for transpose

Implement a VECTOR_IDX_INTO.

There are a lot of DATA(v)[i] littered around, this can be abstracted.

This also adds symmetry with the matrix implementation.