Teo Susnjak

Influences on regression testing strategies in agile software development environments

Regression testing is a well-established practice in software development, but in recent years it has seen a change of status and emphasis with the increasing popularity of agile methods, which stress the central role of regression testing in maintaining software quality. The objectives of this article are to investigate regression testing strategies in agile development teams and identify the factors that can influence the adoption and implementation of this practice. We have used a mixed methods approach to our research, beginning with an analysis of the literature to identify research themes related to the adoption of regression testing techniques under agile methodologies, from which we developed an analytical framework for the study. This was followed by three exploratory case studies that we used to exercise the main elements of the framework, develop some key themes of interest, and devise a questionnaire for the final stage of the study, an on-line survey to explore the main issues identified in the case studies across different contexts. Within our specific sample, our results suggest that organizational maturity is a key factor in effective regression testing practices and that the adoption of such practices is helped by a coherent testing philosophy and change management processes. We also found that the return on investment in automated regression testing was positive for our respondents and that adopting these practices in the context of agile methods had been a relatively painless process for the organizations in our survey. We conclude that investing in regression testing tools and processes is likely to be beneficial for organizations. However, further work is needed in assessing how organizational culture impacts on the quality process and the financial outcomes for commercial software development organizations.

Adaptive cascade of boosted ensembles for face detection in concept drift

We propose an adaptive learning algorithm for cascades of boosted ensembles that is designed to handle the problem of concept drift in nonstationary environments. The goal was to create a real-time adaptive algorithm for dynamic environments that exhibit varying degrees of drift in high-volume streaming data. This we achieved using a hybrid of detect-and-retrain and constant-update approaches. The uniqueness of our method is found in two aspects of our framework. The first is the manner in which individual weak classifiers within each cascade layer of an ensemble are clustered during training and assigned a competence value. Secondly, the idea of learning optimal cascade-layer thresholds during runtime, which enables rapid adaptation to dynamic environments. The proposed adaptive learning method was applied to a binary-class problem with rare-event detection characteristics. For this, we chose the domain of face detection and demonstrate experimentally the ability of our algorithm to achieve an effective trade-off between accuracy and speed of adaptations in dense data streams with unknown rates of change.

A modular approach to training cascades of boosted ensembles

Building on the ideas of Viola-Jones [1] we present a framework for training cascades of boosted ensembles (CoBE) which introduces further modularity and tractability to the training process. It addresses the challenges faced by CoBE frameworks such as protracted runtimes, slow layer convergences and classifier optimization. The framework possesses the ability to bootstrap positive samples and may in turn be extended into the domain of incremental learning. This paper aims to address our frameworks susceptibility to overfitting with possible solutions. Experiments are conducted on face detectors using the bootstrapping of large positive datasets and their accuracy, with respect to overfitting, is examined.

Design from detail

ContextCoderetreats are reflective communities of practice, where participants congregate informally to apply their coding abilities to a clearly defined problem setting with the aim of developing their software design skills. One of these events is the global day of coderetreat (GDCR) involving more than two thousand software developers worldwide. ObjectiveThe GDCR provided an opportunity to explore the ways that the coderetreat activity is perceived by its participants as a medium for reflective practice, and to suggest ways that we can enhance the design of coderetreats for improving both the experience and learning outcomes. MethodWe conducted both quantitative and qualitative surveys from a number of participants in the GDCR to understand how software developers hone their craft in the context of a coderetreat. ResultsOur study indicates that future coderetreats, particularly those that are likely to attract less experienced developers, should consider providing more structural scaffolding to the initial processes of test driven development. ConclusionWe believe that all coderetreats should more explicitly encourage practice and reflection on the four elements of simple design. We suggest a more sustained approach throughout the coderetreat with provision of more structural scaffolding. Accordingly, we have derived a set of recommended practices to make the coderetreat more effective for less experienced developers.

Tuning Fuzzy-Based Hybrid Navigation Systems Using Calibration Maps

We present a novel approach for the tuning and assessment of a cascade of fuzzy logic systems, working cohesively for robot soccer navigation. We generate calibration maps to comprehensively examine the performance of the cascades, allowing for both the visualisation and quantification of the overall system performance. The experiments demonstrate how the proposed method captures the aggregate effect on system’s efficiency of even the slightest changes to the fuzzy rules. It also provides feedback on the mechanics of the fuzzy systems that could be held responsible for any shortcomings. Interestingly, without the aid of the proposed techniques, these minute changes are very difficult, if not impossible to identify through human visual inspection per se. Although the example provided in the paper reflects navigation in the Mirosot league robot soccer scope, the proposed calibration method lends itself amenable to other problem domains where target pursuit and obstacle avoidance behaviours are a necessity. It is also worth-noting that the calibration method can be utilised as a fitness function to a Genetic Algorithm or other optimisation techniques, for a fully-automated calibration. Lastly, we discuss how the calibrated cascade of fuzzy systems neatly integrate with the A* algorithm to produce a hybrid system for near-optimal navigation.

A new ensemble-based cascaded framework for multiclass training with simple weak learners

We present a novel approach to multiclass learning using an ensemblebased cascaded learning framework. By implementing a multiclass cascaded classifier with AdaBoost, we show how detection runtimes are accelerated since only a subset of the ensemble is executed, thus making the classifiers suitable for computer vision applications. We also propose a new multiclass weak learner and demonstrate the frameworks ability to achieve arbitrarily low training errors in conjunction with it. We tested our algorithm against AdaBoost.OC, ECC and M2 multiclass learning methods, on seven benchmark UCI datasets. In our experiments, we found that our framework achieves higher accuracy on five out of seven datasets and displays faster runtime efficiency in all cases.

Adaptive ensemble based learning in non-stationary environments with variable concept drift

The aim of this paper is to present an alternative ensemblebased drift learning method that is applicable to cascaded ensemble classifiers. It is a hybrid of detect-and-retrain and constant-update approaches, thus being equally responsive to both gradual and abrupt concept drifts. It is designed to address the issues of concept forgetting, experienced when altering weights of individual ensembles, as well as realtime adaptability limitations of classifiers that are not always possible with ensemble structure-modifying approaches. The algorithm achieves an effective trade-off between accuracy and speed of adaptations in timeevolving environments with unknown rates of change and is capable of handling large volume data-streams in real-time.

Accelerated classifier training using the PSL cascading structure

This paper addresses the problem of excessively long classifier training times associated with using the Adaboost algorithm within the framework of a cascade of boosted ensembles (CoBE). We present new test results confirming the acceleration of the training phase and the robustness of the Parallel Strong classifier within the same Layer (PSL) training structure recently proposed by [1]. The findings demonstrate a speed up of an order of magnitude over the current training methods without a compromise in accuracy. We also present a modified version of the PSL training structure that further decreases the duration of the training phase while preserving accuracy.

Fast and Smooth Replanning for Navigation in Partially Unknown Terrain: The Hybrid Fuzzy-D*lite Algorithm

This paper presents a hybrid Fuzzy-D*lite algorithm for smoothly navigating robots in an unknown terrain, in real-time. D*lite is a clever optimal, incremental and heuristic search algorithm that is known to be capable of achieving a speed up of one to two orders of magnitude over repeated A* searches. Given a target destination and an incomplete map, it is able to generate a sequence of waypoints for a robot, on the fly, performing course corrections whenever necessary, at a reduced computational time and memory footprint due to its incremental search capability. On the other hand, a cascade of fuzzy systems designed to take advantage of symmetry in the problem domain implements target pursuit and stationary spinning behaviours, for a two-wheeled robot. These reactionary systems calculate the exact steering angle and speed adjustments, enabling the robot to navigate smoothly and fast. We demonstrate how these complementary algorithms can be fused together to achieve smooth and fast continuous re-planning actions in a partially unknown terrain.

Using data-driven and process mining techniques for identifying and characterizing problem gamblers in New Zealand

This paper uses data-driven techniques combined with established theory in order to analyse gambling behavioural patterns of 91 thousand individuals on a real-world fixed-odds gambling dataset in New Zealand. This research uniquely integrates a mixture of process mining, data mining and confirmatory statistical techniques in order to categorise different sub-groups of gamblers, with the explicit motivation of identifying problem gambling behaviours and reporting on the challenges and lessons learned from our case study. We demonstrate how techniques from various disciplines can be combined in order to gain insight into the behavioural patterns exhibited by different types of gamblers, as well as provide assurances of the correctness of our approach and findings. A highlight of this case study is both the methodology which demonstrates how such a combination of techniques provides a rich set of effective tools to undertake an exploratory and open-ended data analysis project that is guided by the process cube concept, as well as the findings themselves which indicate that the contribution that problem gamblers make to the total volume, expenditure and revenue is higher than previous studies have maintained.