A simulation framework for supervised learning data. The functionalities are specifically designed to let the user a maximum degrees of freedom, to ultimately fulfill the research purpose. Furthermore, feature importances of the simulation can be created on a local and a global level. This is particular interesting, for instance, to benchmark feature selection algorithms.
Simulating Supervised Learning Data
xypy.Xy() you can convienently simulate supervised learning data, e.g. regression and classification problems.
The simulation can be very specific, since there are many degrees of freedom for the user. For instance, the functional
shape of the nonlinearity is user-defined as well. Interactions can be formed and (co)variances altered. For a more
specific motivation you can visit our blog.
I have adapted this package from my R version, which you can check out here.
You can convienently install the package via PYPI with the following command.
pip install xypy
There is an example test script on my GitHub, which will you get started with the simulation.
You can simulate regression and classification data with interactions and a user-specified non-linearity. With
stn argument you can alter the signal to noise ratio of your simulation. I strongly encourage you to
read this blog post,
where I've analyzed OLS coefficients with different signal to noise ratios.
# load the library from xypy import Xy # simulate regression data my_sim = Xy(n = 1000, numvars = [10,10], catvars = [3, 2], noisevars = 50, stn = 100.0) # get a glimpse of the simulation my_sim # plot the true underlying effects my_sim.plot() # extract the data X, y = my_sim.data # extract the true underlying model weights my_sim.coef_
You can extract a feature importance of your simulation. For instance, to benchmark feature selection algorithms. You can read up on a small benchmark I did with this feature on our blog. You can perform the same analysis easily in Python as well.
# Feature Importance my_sim.varimp()
Feel free to contact me with input and ideas.
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.