Installation

If you want to generate the Python package without installing it :

maturin build (--release)

You'll find a file named libgcmpyo3.dylib in target/{debug,release}. Just rename the file into gcmpyo3.so to use it. Otherwise, you can pip install the wheel file located in the target/wheels folder.

Use maturin develop to compile + install the library.

Testing

To make sure the module is correctly loaded, run

>>> import gcmpyo3
>>> gcmpyo3.test()
The module is loaded correctly

Summary of variable names

Parameters

$\lambda$ : $\ell_2$ penalization of logistic regression

$\beta$ : inverse temperature in pseudo-bayes

$\alpha$ : sampling ratio

$\delta$ : variance of the Gaussian noise

$\gamma$ : ratio between student and teacher dimensions

$\kappa_1, \kappa_{\star}$ : parameters of the student covariance in the random feature case