hanshengjiang / npmle_git

A Python package for computing NPMLE of mixture of regression

NPMLE: Nonparametric Maximum Likelihood Estimation of Mixture of Regression

Overview

This repository contains reproduction code and data for numerical implementation in the paper A Nonparametric Maximum Likelihood Approach to Mixture of Regression. Most scripts are in Python except one R script. These files have been developed and tested in Python version 3.7.4 and R version 3.6.1.

Folder

• scripts/: (1) scripts (named '*_lib.py') that implement the NPMLE procedures, including set-ups for simulation and plotting; (2) scripts (named 'run_*.py') that carry out simulations and real data analysis
• data/: various results stored in .csv files
• pics/: various visualization results
• real_data/: data .csv files for real data analysis

Usage

Single run

Each script starting with 'run_' in scripts/ is used for one run of a certain numerical experiment. The users can directly run the script in their own IDE but running the script from the terminal is recommended. For example,

python run_real_data_gdp.py

or

python run_simulation.py 

When running scripts in terminal, certain command-line arguments can be passed to the script in order to switch among different simulation settings. For example, for script run_simulation.py one can specify a set of command line arguments as below, which represent that scale parameter = 0.5, number of data points = 500, component coefficients setup = type 1, whether to run cross-validation = yes, and the granularity of cross-validation = 0.01 respectively.

python run_simulation.py 0.5 500 1 yes 0.01

The users can consult the configuration part at the beginning of the scripts to find all arguments that are available.

Multiple run

There are also scripts in scripts/ that can run multiple simulations with a one-line command. For example, one can use the script run_multiple_simulations.py to run all simulation examples for discrete cases by adding two command-line arguments as follows.

python run_multiple_simulations.py run_simulation.py discrete_cv

There are also a few other command-line arguments available for this script, which are specified in documentation.

License

All content in this repository is licensed under the MIT license. Comments and suggestions are welcome!