This repository contains the code to reproduce the experiments in our paper: Linear predictor on linearly-generated data with missing values: non consistency and solutions.

The file environment_lpm.yml indicates the packages required to run the code as well as the versions that were used.

The file ground_truth.py contains the parameter generation, data generation and Bayes rate functions.

The file estimators.py contains all the classes of estimators used.

The file learning_curves.py contains the code which runs the experiments.

The file launch_experiment takes mixture1, mixture3, or selfmasked_proba as argument. For example python launch_experiment.py mixture1 launches the simulations for mixture1. Change this file if you want to change the values of the parameters tested for the simulations. Upon completion of this script, a csv file is saved that records the performances obtained.

The file plot_curves, plot_MLP_scaling_n, plot_MLP_scaling_q, plot_boxplots contain the code used to plot the figures based on the csv file obtained. plot_curves plots the learning curves.