Muhammad Rezaul Karim

Bi-objective Genetic Search for Release Planning in Support of Themes

Release planning is a mandatory part of incremental and iterative software development. For the decision about which features should be implemented next, the values of features need to be balanced with the effort and readiness of their implementation. Traditional planning looks at the sum of the values of individual and potentially isolated features. As an alternative idea, a theme is a meta-functionality which integrates a number of individual features under a joint umbrella. That way, possible value synergies from offering features in conjunction (theme-related) can be utilized.

Sensitive ants are sensible ants

This paper introduces an approach to evolving computer programs using an Attribute Grammar (AG) extension of Grammatical Evolution (GE) to eliminate ineffective pieces of code with the help of context-sensitive information. The standard Context-Free Grammars (CFGs) used in GE, Genetic Programming (GP) (which uses a special type of CFG with just a single non-terminal) and most other grammar-based system are not well-suited for codifying information about context. AGs, on the other hand, are grammars that contain functional units that can help determine context which, as this paper demonstrates, is key to removing ineffective code. The results presented in this paper indicate that, on a selection of grammars, the prevention of the appearance of ineffective code through the use of context analysis significantly improves the performance of and resistance to code bloat over both standard GE and GP for both Santa Fe Trail (SFT) and Los Altos Hills (LAH) trail version of the ant problem with same amount of energy used.

A new approach to solving 0-1 multiconstraint knapsack problems using attribute grammar with lookahead

In this paper, we introduce a new approach to genotype-phenotype mapping for Grammatical Evolution (GE) using an attribute grammar (AG) to solve 0-1 multiconstraint knapsack problems. Previous work on AGs dealt with constraint violations through repeated remapping of non-terminals, which generated many introns, thus decreasing the power of the evolutionary search. Our approach incorporates a form of lookahead into the mapping process using AG to focus only on feasible solutions and so avoid repeated remapping and introns. The results presented in this paper show that the proposed approach is capable of obtaining high quality solutions for the tested problem instances using fewer evaluations than existing methods.

Who Should Take This Task?: Dynamic Decision Support for Crowd Workers

Context: The success of crowdsourced software development (CSD) depends on a large crowd of trustworthy software workers who are registering and submitting for their interested tasks in exchange of financial gains. Preliminary analysis on software worker behaviors reveals an alarming task-quitting rate of 82.9%. Goal: The objective of this study is to empirically investigate worker decision factors and provide better decision support in order to improve the success and efficiency of CSD. Method: We propose a novel problem formulation, DCW-DS, and an analytics-based decision support methodology to guide workers in acceptance of offered development tasks. DCS-DS is evaluated using more than one years real-world data from TopCoder, the leading CSD platform. Results: Applying Random Forest based machine learning with dynamic updates, we can predict a worker as being a likely quitter with 99% average precision and 99% average recall accuracy. Similarly, we achieved 78% average precision and 88% average recall for the worker winner class. For workers just following the top three task recommendations, we have shown that the average quitting rate goes down below 6%. Conclusions: In total, the proposed method can be used to improve total success rate as well as reduce quitting rate of tasks performed.

An empirical investigation of single-objective and multiobjective evolutionary algorithms for developer's assignment to bugs

In this paper, the modeling of developers’ assignment to bugs (DAB) is studied. The problem is modeled both as a single objective (minimize bug fix time) and as a bi‐objective (minimize bug fix time and cost) combinatorial optimization problem. Two models of developer assignment are considered where in the first model a single developer is assigned per bug (single developer model), while in the second model a single developer is assigned for each competency area of a bug (individual competency model). The latter model is proposed in this paper. For the single developer model, GA@DAB, an existing genetic algorithm‐based approach, is extended to support precedence among bugs. For the individual competency model of DAB, one genetic algorithm‐based approach (Competence@DAB) and one nondominated sorting genetic algorithm II‐based approach (CompetenceMulti2@DAB ) are proposed to generate solutions minimizing time and minimizing both time and cost, respectively. The performance of the proposed approaches was evaluated for 2040 bugs of 19 open‐source milestone projects from the Eclipse platform. Our results and analysis show that the proposed individual competency model is far better than the single developer model, with average bug fix time reduction of 39.7% across all projects. Copyright

Comparative analysis of predictive techniques for release readiness classification

Context: A software release is the deployment of a version of an evolving software product. Product managers are typically responsible for deciding the release content, time frame, price, and quality of the release. Due to all the dynamic changes in the project and process parameters, the decision is highly complex and of high impact.Objective: This paper has two objectives: i) Comparative analysis of predictive techniques in classifying an ongoing release in terms of its expected release readiness., and ii) Comparative analysis between regular and ensemble classifiers to classify an ongoing release in terms of its expected release readiness.Methodology: We use machine learning classifiers to predict release readiness. We analyzed three OSS projects under Apache Software Foundation from JIRA issue repository. As a retrospective study, we covered a period of 70 months, 85 releases and 1696 issues. We monitored eight established variables to train classifiers in order to predict whether releases will be ready versus non-ready. Predictive performance of different classifiers was compared by measuring precision, recall, F-measure, balanced accuracy, and area under the ROC curve (AUC).Results: Comparative analysis among nine classifiers revealed that ensemble classifiers significantly outperform regular classifiers. Balancing precision and recall, Random Forrest and BaggedADABoost were the two best performers in total, while Naïve Bayes performed best among just the regular classifiers.

Attributed Grammatical Evolution with Lookahead for the Multiple Knapsack Problem

This paper presents an Attribute Grammar with Lookahead (AG+LA) approach, a technique to solve heavily constrained Multiple Knapsack Problem. This approach incorporates a form of lookahead into the mapping process of Grammatical Evolution (GE) using Attribute Grammar (AG) to focus only on feasible solutions, thereby avoiding issues such as repeated remapping and introns, both of which are limitations of previous approaches based on AG. We also present AG+LAE (AG+LA with an efficiency measure to bias the search towards the most efficient, i.e., best value, objects), the successor of AG+LA where a biasing process is incorporated using problem specific knowledge to significantly improve the performance of its predecessor, both in terms of the number of evaluations required and the quality of solutions obtained. Degenerate code, as used in DNA, is code that uses redundancy, so that different codons can represent the same thing. Many developmental systems, such as GE, use a degenerate encoding to help promote neutral mutations, that is, minor genetic changes that do not result in a phenotypic change. While early work in GE suggested that some level of degeneracy was important, it does come at the cost of increasing the size of the search space. Duplicate Elimination techniques, as opposed to degenerate encoding, are employed in decoder-based Evolutionary Algorithms to ensure that the newly generated solutions are not already contained in the current population. The results and analysis show that it is crucial to incorporate duplicate elimination to improve the performance of both approaches, while the reduced level of degeneracy is crucial only for AG+LA.

Degeneracy Reduction or Duplicate Elimination? An Analysis on the Performance of Attributed Grammatical Evolution with Lookahead to Solve the Multiple Knapsack Problem

This paper analyzes the impact of having degenerate code and duplicate elimination in an attribute grammar with lookahead (AG+LA) approach, a recently proposed mapping process for Grammatical Evolution (GE) using attribute grammar (AG) with a lookahead feature to solve heavily constrained multiple knapsack problems (MKP). Degenerate code, as used in DNA, is code in which different codons can represent the same thing. Many developmental systems, such as (GE), use a degenerate encoding to help promote neutral mutations, that is, minor genetic changes that do not result in a phenotypic change. Early work on GE suggested that at least some level of degeneracy has a significant impact on the quality of search when compared to the system with none. Duplicate elimination techniques, as opposed to degenerate encoding, are employed in decoder-based Evolutionary Algorithms (EAs) to ensure that the newly generated solutions are not already contained in the current population. The results and analysis show that it is crucial to incorporate duplicate elimination to improve the performance of AG+LA. Reducing level of degeneracy is also important to improve search performance, specially for the large instances of the MKP.

Applying data analytics towards optimized issue management: an industrial case study

This document describes our experience of applying data analytics at Plexina, a leading IT company working in the healthcare domain. The main goal of the project was to identify factors currently affecting issue management and to make analytics based suggestions for optimizing the process. Various statistical and machine learning techniques were applied on a data set extracted from six releases of Plexina, containing more than 666 issues. Statistical techniques successfully identified the various factors that leads to estimation inaccuracy related to issues as well as identified the hidden relationships existing among various variables. The employed predictive analytic models was also successful to some extent, in predicting effort estimation related inaccuracy associated with the issues. The insights provided by the entire data analytics study can be of great help to product managers or the developers to make more informed decisions. In addition, the guidelines presented in this paper based on the lessons learnt can be applied to other data analytics and academia-industry collaboration project.

On improving grammatical evolution performance in symbolic regression with attribute grammar

This paper shows how attribute grammar (AG) can be used with Grammatical Evolution (GE) to avoid invalidators in the symbolic regression solutions generated by GE. In this paper, we also show how interval arithmetic can be implemented with AG to avoid selection of certain arithmetic operators or transcendental functions, whenever necessary to avoid infinite output bounds in the solutions. Results and analysis demonstrate that with the proposed extensions, GE shows significantly less overfitting than standard GE and Kozas GP, on the tested symbolic regression problems.