Abstract Contributor turnover is a major challenge in open-source software (OSS) projects, affecting project sustainability, quality, and many other dimensions of project performance. Although turnover has been widely studied in the OSS context, the importance of sustained contribution periods to long-term contributor survival remains poorly understood. In this context, contributor survival refers to the probability of sustained activity over time. We investigate the effect of contribution streaks, defined as consecutive months of contribution, on long-term survival, and analyze what kind of engagement, at different levels of ownership, predicts short-term contribution streaks. Using five years of data from 14 open-source projects, we applied landmark analysis to understand long-term survival based on peak streak lengths and logistic regression to identify the activities supporting short-term streaks. In both cases, we identified a three-month contribution streak to be an important threshold for sustained contributions. We found that development and communication related activities across ownership levels are important predictors of whether contributors continue their streaks, even though the relative importance shifts over time. In addition, we found that prior experience plays a stronger role early on, but its effect diminishes during later streak months. These findings suggest that contribution streaks are important predictors of long-term survival and provide practical insights for open-source practitioners on how to improve the retention of new and existing contributors through supporting short-term contribution streaks.

Data science projects have become commonplace over the last decade. During this time, the practices of running such projects, together with the tools used to run them, have evolved considerably. Furthermore, there are various studies on data science workflows and data science project teams. However, studies looking into both workflows and teams are still scarce and comprehensive works to build a holistic view do not exist. This study bases on a prior case study on roles and processes in data science. The goal here is to create a deeper understanding of data science projects and development processes. We conducted a survey targeted at experts working in the field of data science (n=50) to understand data science projects’ team structure, roles in the teams, utilized project management practices and the challenges in data science work. Results show little difference between big data projects and other data science. The found differences, however, give pointers for future research on how agile data science projects are, and how important is the role of supporting project management personnel. The current study is work in progress and attempts to spark discussion and new research directions.

Software development projects have increasingly been adopting new practices, such as continuous delivery and deployment to enable rapid delivery of new features to end users. Tools that are commonly utilized with these practices generate a vast amount of data concerning various development events. Analysis of the data provides a lightweight data driven view on the software process. We present an efficient way of visualizing software process data to provide a good overall view on the features and potential problems of the process. We use the visualization in a case project that has become more agile by applying continuous integration and delivery together with development and infrastructure automation. We compare data visualizations with information gathered from the development team and describe how the evolution can be understood through our visualizations. The case project is a good example of how a change from a traditional long cycle development to a rapid cycle DevOps culture can actually be made in a few years. However, the results show that the team has to focus on the process improvement continuously in order to maintain continuous delivery all the time. As the main contribution, we present a lightweight way to software process visualization. Moreover, we discuss how such a heuristic can be used to track the characteristics of the target process.

Data warehousing is a process of integrating multiple data sources into one for, e.g., reporting purposes. An emerging modeling technique for this is the data vault method. The use of data vault creates many structurally similar data processing modifications in the transform phase of ETL work. Is it possible to automate the creation of transformations? Based on our study, the answer is mostly affirmative. Data vault modeling creates certain constraints to data warehouse entities. These model constraints and data vault table populating principles can be used to generate transformation code. Based on the original relational database model and data flow metadata we can gather populating principles. These can then be used to create general templates for each entity. Nevertheless, we need to note that the use of data flow metadata can be only partially automated and includes the only manual work phases in the process. In the end we can generate the actual transformation code automatically. In this paper, we carefully describe the creation of automation procedure and analyze the practical problems based on our experiences on PL/SQL proof of concept implementation. To the best of our knowledge, similar has not yet been described in the scientific literature.

While the users of completed applications are heavily moving from desktop to the web browser, the majority of developers are still working with desktop IDEs such as Eclipse or Visual Studio. In contrast to professional installable IDEs, current web-based code editors are simple text editors with extra features. They usually understand lexical syntax and can do highlighting and indenting, but lack many of the features seen in modern desktop editors. In this paper, we present CoRED, a browser-based collaborative real-time code editor for Java applications. CoRED is a complete Java editor with error checking and automatic code generation capabilities, extended with some features commonly associated with social media. As a proof of the concept, we have extended CoRED to support Java based Vaadin framework for web applications. Moreover, CoRED can be used either as a stand-alone version or as a component of any other software. It is already used as a part of browser based Arvue IDE.

Methods for learning decision rules are being successfully applied to many problem domains, especially where understanding and interpretation of the learned model is necessary. In many real life problems, we would like to predict multiple related (nominal or numeric) target attributes simultaneously. Methods for learning rules that predict multiple targets at once already exist, but are unfortunately based on the covering algorithm, which is not very well suited for regression problems. A better solution for regression problems may be a rule ensemble approach that transcribes an ensemble of decision trees into a large collection of rules. An optimization procedure is then used for selecting the best (and much smaller) subset of these rules, and to determine their weights. Using the rule ensembles approach we have developed a new system for learning rule ensembles for multi-target regression problems. The newly developed method was extensively evaluated and the results show that the accuracy of multi-target regression rule ensembles is better than the accuracy of multi-target regression trees, but somewhat worse than the accuracy of multi-target random forests. The rules are significantly more concise than random forests, and it is also possible to create very small rule sets that are still comparable in accuracy to single regression trees.

This paper studies how useful the standard 2-norm regularized SVM is in approximating the 1-norm SVM problem. To this end, we examine a general method that is based on iteratively re-weighting the features and solving a 2-norm optimization problem. The convergence rate of this method is unknown. Previous work indicates that it might require an excessive number of iterations. We study how well we can do with just a small number of iterations. In theory the convergence rate is fast, except for coordinates of the current solution that are close to zero. Our empirical experiments confirm this. In many problems with irrelevant features, already one iteration is often enough to produce accuracy as good as or better than that of the 1-norm SVM. Hence, it seems that in these problems we do not need to converge to the 1-norm SVM solution near zero values. The benefit of this approach is that we can build something similar to the 1-norm regularized solver based on any 2-norm regularized solver. This is quick to implement and the solution inherits the good qualities of the solver such as scalability and stability.