knockpy 1.0.0

Knockoffs for variable selection

Knockpy

A python implementation of the knockoffs framework for variable selection. See https://amspector100.github.io/knockpy/ for detailed documentation and tutorials.

Installation

To install knockpy, first install choldate:

pip install git+git://github.com/jcrudy/choldate.git

Then, install knockpy using pip:

pip install knockpy

If the installation fails on your system, please reach out to me and I'll try to help.

To use the (optional) kpytorch submodule, you will need to install pytorch.

Quickstart

Given a data-matrix X and a response vector y, knockpy makes it easy to use knockoffs to perform variable selection using a wide variety of machine learning algorithms (also known as "feature statistic") and types of knockoffs. One quick example is shown below, where we use the cross-validated lasso to assign variable importances to the features and knockoffs.

    import knockpy as kpy
    from knockpy.knockoff_filter import KnockoffFilter

    # Generate synthetic data from a Gaussian linear model
    data_gen_process = kpy.dgp.DGP()
    data_gen_process.sample_data(
        n=1500, # Number of datapoints
        p=500, # Dimensionality
        sparsity=0.1,
        x_dist='gaussian',
    )
    X = data_gen_process.X
    y = data_gen_process.y
    Sigma=data_gen_process.Sigma

    # Run model-X knockoffs
    kfilter = KnockoffFilter(
        fstat='lasso',
        ksampler='gaussian',
    )
    rejections = kfilter.forward(X=X, y=y, Sigma=Sigma)

Most importantly, knockpy is built to be modular, so researchers and analysts can easily layer functionality on top of it.

To run tests

• To run all tests, run python3 -m pytest
• To run a specific label, run pytest -v -m {label}.
• To select all labels except a particular one, run pytest -v -m "not {label}" (with the quotes).
• To run a specific file, try pytest test/{file_name}.py. To run a specific test within the file, run pytest test/{file_name}.py::classname::test_method. You also don't have to specify the exact test_method, you get the idea.
• To run a test with profiling, try python3 -m pytest {path} --profile. This should generate a set of .prof files in prof/. Then you can run snakeviz filename.prof to visualize the output. There are also more flags/options for outputs in the command line command.
• Alternatively, cprofilev is much better. To run cprofilev, copy and paste the test to proftest/* and then run python3 -m cprofilev proftest/test_name.py.