HMLI: Heuristic Machine-learning Imputation

`HMIL`_is univeriate time series imputation method based on Heuristic search and machine learning.

Features

• Genetic algorithm to find optimum dependent variables
• Support Vector Machine for regression and prediction

Using HMLI

Edit data file location and evolutationary parameter in hmli.ini:

Default parameter values:

• number of prediction observations: 12
• number of genes per chromosome: 6
• number of chromosomes per population: 10
• number of populations: 100
• cutoff rate of bad chromosome: 0.5
• cutoff correlation coefficient: 0.97
• random seed list: [1, 2, 3]
• missing rate list: [0.1, 0.4, 0.7]
• hdf5 data file: data/mei.h5
• result file: output/finalRes1.pickle

Execute the main file:

..code-block:: bash

$python main.py

It will execute parallel process and consume CPU resource a lot.

Sample data is OECD MEI monthly series. The total number series in the file is 5,411.