samuellees / gpecoc

Genetic Programming based Error Correcting Output Codes

GPECOC

Genetic Programming based Error Correcting Output Codes

This is the implementation for paper: A Novel Error-Correcting Output Code Algorithm Based on Genetic Programming

Acknowledgement

• Codes about Genetic Programming is modified from Pyevolve

• Codes about Output-Code-Classifier is modified from scikit-learn 0.18

Environment

• Windows 10 64 bit

• python 2.7

• Excel

  Enable the macro in excel, so it can extract result from file automatically.

• scikit-learn 0.18

  Anaconda is strongly recommended, run the following command in the Powershell, all necessary python packages for this project will be installed:

  conda install scikit-learn==0.18

Dataset

• Data format

  Data should be put into the folder gpecoc/data. Each dataset should be divided into "dataname_train.data", "datanam_test.data" and "dataname_validation.data". In the sub-dataset, each column is a sample, the first line represents the labels, the rest are feature space. There are two examples datasets in the folder gpecoc/data. Please note that invalid sample, such as value missed, will cause errors.

• Data processing

  Feature Selection and Scaling will be done automatically.

Multi-processing Mode

• Config

  Firstly, make configuration in Configurations.py. In Multi-processing mode, you need not to pay attention to 'dataName' and 'aimFolder'.

• Run the following command

  It will traversal all datasets given by the main function in _ParallelRunner.py, each dataset will be run for 10 times.

  python _ParallelRunner.py

• Analyze result

  All result infos will be written into the folder. For example, if you set version = "8.80", result infos will be found in ($root_path)/Results/8.80/

Single-processing Mode

This is useful when you want to debug.

• Config

  Make configuration in Configurations.py. In Single-processing mode, 'dataName' and 'aimFolder' should also be set.

• Run the following command

  It will do training and testing on the dataset given in the Configurations.py.

  python AGpEcocStart.py

• Parse and analyze result

  In this Mode, part of the result will be printed on the terminal. You can find all result in the Results folder. But, there will be no automatic analyzing.

Attention

Make sure the version folder is not exit every time you run it. Or errors will happen in the following way:

• In Single-processing Mode, the result in the terminal is right, but the result written ino the file might be wrong.
• In Multi-processing Mode, the result could not be read and parsed correctly.

A suggestion is to change the 'version' in the Configurations.py every time you run it.