/Reinforced-Mixture-Learning

Primary LanguageJupyter NotebookMIT LicenseMIT

Reinforced-Mixture-Learning

This repository is the official implementation of the paper Reinforced-Mixture-Learning submitted to Neural Networks.

Requirements

  • Python version: Python 3.6.8 :: Anaconda custom (64-bit)

Main packages for the proposed estimator

  • numpy==1.19.5
  • pandas==1.1.5
  • sklearn==0.24.2
  • torch==1.8.0 (cpu)

Additional packages for experiments

  • os
  • sys
  • random
  • matplotlib

Hardware

  • Precision Tower 7910
  • CPU：Intel(R) Xeon(R) CPU E5-2640 v4 @ 2.40GHz （2 physical CPUs with 10 cores each）

Reproduce the results of experiments

We provide an example code in example.ipynb.

For synthetic data analysis, we consider Gaussian mixture settings and general settings.

fig

You can run gmm.py and model_free.py in the synthetic_data_analysis to get the results in results. The files named likefigure_xx.py are used to generate the figures in the paper. Each file is self-contained.

For real data analysis, we apply our method to three UCI benchmark datasets summarized in the table below.

You can run each {dataset_name}.py in the real_data_analysis to get the results in results. The files named likefigure_xx.py are used to generate the figures in the paper. Each file is self-contained.