ginfung/constraintanalysis

constraintanalysis's Issues

NSGA II Applied

Pruning based on NSGA-II

Convergence rate

Convergence rate is a very important point in comparison. Visualize the converge process

Efficiency of Basic IBEA

It seems that the environment selection will take a very long time. Algorithm simplification is needed.

cliffdelta

https://gist.github.com/timm/a6e759eb7d9b5f05b468

Communication between matlab and java

Anyone know how to call a java object's parent method in Matlab.
I've tried many times, but fail.

Disucssion on applying NSGA II

Most time-consuming part: "sorting"
Even though pruning probability has been reduced, the pruning process is taken a large part of execution time.

More details for IBEA

IBEA- Indicator-Based Evolutionary Algorithm
More details can be found at
Zitzler, Eckart, and Simon Künzli. "Indicator-based selection in multiobjective search." Parallel Problem Solving from Nature-PPSN VIII. Springer Berlin Heidelberg, 2004.

Pruning bug fix

Fix a bug in pruning. Test set has been passed.

Applying SPLOT parser

SPLOT parser is a java lib provided by http://gsd.uwaterloo.ca/

Scalable multiply objectives result add

Finish adding the comparison between DE and IBEA.
DE runs faster than IBEA
IBEA cannot converge at the expected way.
IBEA can generate better spread.

A12 statistic

Add a12 statistic.
Reference: Vargha, András, and Harold D. Delaney. "A critique and improvement of the CL common language effect size statistics of McGraw and Wong." Journal of Educational and Behavioral Statistics 25.2 (2000): 101-132.

Converage for IBEA

Results shows that IBEA cannot converge. But indeed it can generate more spread.
Reason may be:
crossover in the algorithm.
mutation in GA

add project introduction in readme.md

there are some format problems in this introduction.

Bibliography collected

All the bibliography have been collected and list in the reference.bib

Diversity Preservation issus

Sometimes the DE will converge at a incorrect frontier. The main reason for this, according to my observation, is the lack of data diversity. It cannot generate new mutation, that is, the result converges too early.

another map: continuous space to binary space

Reference: Pampara, Gary, Andries Petrus Engelbrecht, and Nelis Franken. "Binary differential evolution." Evolutionary Computation, 2006. CEC 2006. IEEE Congress on. IEEE, 2006.

testing for some feature model

feature model listed in #10 (1,2) as well as model No3,5 in SPLOT passed

Bounds error in xtile

BUG: xtile may throw a bug.
e,g. lst = [1,1,1,1]

Classic SXFM models

Some classic SXFM models to test the system.

1. First one, No.1 in http://splot-research.org/
2. Web Portal No.523 in http://splot-research.org/
3. [huge constraints] Car selection No.592 in http://splot-research.org/

Current result showed in report proj1a

The current result will be presented in the report proj1a.

so-called "improve the frontier set"

what is so-called "move forward to the Pareto-frontier set"?

reference on binary torunament selection

http://cstheory.stackexchange.com/questions/14758/tournament-selection-in-genetic-algorithms

New branch for large feature model testing

Created a branch for testing large model.

Validity and search space

Since many of feature models are valid in SXFM (at least one solution that does not violate any constraints), should we lower the search space--only searching the valid solution?
It is not easy to find a balance between constraint maintenance and other objectives.
[inspired by the blending and arbitration in CSC584]

More feature attribute assigned

Costs, Used_before and defects have been assigned as additional attributes for the feature.
REFERENCE:
Sayyad, Abdel Salam, Tim Menzies, and Hany Ammar. "On the value of user preferences in search-based software engineering: A case study in software product lines." Software engineering (ICSE), 2013 35th international conference on. IEEE, 2013.

Atom feature

An idea about atom feature sets.
In actual optimization, some features should be regarded as one feature. That is, assuming that constraint satisfaction is first thing, some features should share the same value. All these features should put together as one atom feature.

Moderate feature Model

Web portal is a moderate feature model, which can be found at http://gsd.uwaterloo.ca:8088/SPLOT/models/REAL-FM-3.xml

Add three attributes for each feature: cost, usedbefore and defects

cost in [5,15]
usedbefore is boolean value distribute uniformly
and defect in [0,10]

also, defects should be 0 when usedbefore = false

Test Entrance

Add unique test entrance.
Unify the test initial data.

Modified the feature attributes

Try to modified the attributes for some features so that the result can be more reasonable.

Indicator normalization

The indicators should be normalized when calculating. Bug has been fixed.