Feature Decay Algorithms Optimization with Genetic Algorithms
Copyright (c) 2014, Ergun Bicici, ergunbicici@yahoo.com
Citation:
Ergun Bicici and Deniz Yuret, “Optimizing Instance Selection for Statistical Machine Translation with Feature Decay Algorithms”, IEEE/ACM Transactions On Audio, Speech, and Language Processing (TASLP), 2014.
Feature Decay Algorithms (FDA) is developed as part of my PhD thesis:
Ergun Biçici. The Regression Model of Machine Translation. PhD thesis, Koç University, 2011. Note: Supervisor: Deniz Yuret.
The program is using genetic algorithms (GA) for optimizing FDA. It uses inspyred package for optimization.
TO RUN: (1) Install the inspyred package:
unzip inspyred.zip
bash build_distribution.sh
python setup.py install
(2) Prepare a config file for the optimization run. You can use the provided sample config file for this purpose:
fda.optimization.config
(3) Run an optimized fda:
python runoptFDA.py fda.optimization.config [numproc]
numproc is the optional parameter for the number of processors. Default value is 24.
Sample run output is given in file samplerun.output and below:
optimization settings: ('../fda/fda', 'data/train.en', 'data/train.de', 'data/dev.en', 'data/dev.de', 2, 92681) Optimization took: 5476.26002908 ../fda/fda -v1 -t92681 -d1.0 -c1.0089124001 -s0.886387647059 -i2.16465456164 -l1.89379071178 -n2 -o fda.optimization.config.selection data/train.en data/dev.en data/train.de data/dev.de FDA5 took: 37.7974550724 Overall time: 5514.05750203