Code Repositories Research Articles

Deep Graph Matching and Searching for Semantic Code Retrieval

Code retrieval is to find the code snippet from a large corpus of source code repositories that highly matches the query of natural language description. Recent work mainly uses natural language processing techniques to process both query texts (i.e., human natural language) and code snippets (i.e., machine programming language), however, neglecting the deep structured features of query texts and source codes, both of which contain rich semantic information. In this article, we propose an end-to-end deep graph matching and searching (DGMS) model based on graph neural networks for the task of semantic code retrieval. To this end, we first represent both natural language query texts and programming language code snippets with the unified graph-structured data, and then use the proposed graph matching and searching model to retrieve the best matching code snippet. In particular, DGMS not only captures more structural information for individual query texts or code snippets, but also learns the fine-grained similarity between them by cross-attention based semantic matching operations. We evaluate the proposed DGMS model on two public code retrieval datasets with two representative programming languages (i.e., Java and Python). Experiment results demonstrate that DGMS significantly outperforms state-of-the-art baseline models by a large margin on both datasets. Moreover, our extensive ablation studies systematically investigate and illustrate the impact of each part of DGMS.

Combining weakly and strongly supervised learning improves strong supervision in Gleason pattern classification

BackgroundOne challenge to train deep convolutional neural network (CNNs) models with whole slide images (WSIs) is providing the required large number of costly, manually annotated image regions. Strategies to alleviate the scarcity of annotated data include: using transfer learning, data augmentation and training the models with less expensive image-level annotations (weakly-supervised learning). However, it is not clear how to combine the use of transfer learning in a CNN model when different data sources are available for training or how to leverage from the combination of large amounts of weakly annotated images with a set of local region annotations. This paper aims to evaluate CNN training strategies based on transfer learning to leverage the combination of weak and strong annotations in heterogeneous data sources. The trade-off between classification performance and annotation effort is explored by evaluating a CNN that learns from strong labels (region annotations) and is later fine-tuned on a dataset with less expensive weak (image-level) labels.ResultsAs expected, the model performance on strongly annotated data steadily increases as the percentage of strong annotations that are used increases, reaching a performance comparable to pathologists (kappa = 0.691 pm 0.02). Nevertheless, the performance sharply decreases when applied for the WSI classification scenario with kappa = 0.307 pm 0.133. Moreover, it only provides a lower performance regardless of the number of annotations used. The model performance increases when fine-tuning the model for the task of Gleason scoring with the weak WSI labels kappa = 0.528 pm 0.05.ConclusionCombining weak and strong supervision improves strong supervision in classification of Gleason patterns using tissue microarrays (TMA) and WSI regions. Our results contribute very good strategies for training CNN models combining few annotated data and heterogeneous data sources. The performance increases in the controlled TMA scenario with the number of annotations used to train the model. Nevertheless, the performance is hindered when the trained TMA model is applied directly to the more challenging WSI classification problem. This demonstrates that a good pre-trained model for prostate cancer TMA image classification may lead to the best downstream model if fine-tuned on the WSI target dataset. We have made available the source code repository for reproducing the experiments in the paper: https://github.com/ilmaro8/Digital_Pathology_Transfer_Learning

Accounting for Geographic Basis Risk in Heat Index Insurance: How Spatial Interpolation Can Reduce the Cost of Risk

AbstractExtreme heat events cause periodic damage to crop yields and may pose a threat to the income of farmers. Weather index insurance provides payouts to farmers in the case of measurable weather extremes to keep production going. However, its viability depends crucially on the accuracy of local weather indices to predict yield damages from adverse weather conditions. So far, extreme heat indices are poorly represented in weather index insurance. In this study, we construct indices of extreme heat using observations at the nearest weather station and estimates for each county using three interpolation techniques: inverse-distance weighting, ordinary kriging, and regression kriging. Applying these indices to insurance against heat damage to corn in Illinois and Iowa, we show that heat index insurance reduces relative risk premiums by 27%–29% and that interpolated indices outperform the nearest-neighbor index by around 2%–3% in terms of relative risk reduction. Further, we find that the advantage of interpolation over a nearest-neighbor index in terms of relative risk reduction increases as the sample of weather stations is reduced. These findings suggest that heat index insurance can work even when weather data are spatially sparse, which delivers important implications for insurance practice and policy makers. Further, our public code repository provides a rich toolbox of methods to be used for other perils, crops, and regions. Our results are therefore not only replicable but also constitute a cornerstone for projects to come.

On the Use of ‘Glyphmaps’ for Analysing the Scale and Temporal Spread of COVID-19 Reported Cases

Recent analysis of area-level COVID-19 cases data attempts to grapple with a challenge familiar to geovisualization: how to capture the development of the virus, whilst supporting analysis across geographic areas? We present several glyphmap designs for addressing this challenge applied to local authority data in England whereby charts displaying multiple aspects related to the pandemic are given a geographic arrangement. These graphics are visually complex, with clutter, occlusion and salience bias an inevitable consequence. We develop a framework for describing and validating the graphics against data and design requirements. Together with an observational data analysis, this framework is used to evaluate our designs, relating them to particular data analysis needs based on the usefulness of the structure they expose. Our designs, documented in an accompanying code repository, attend to common difficulties in geovisualization design and could transfer to contexts outside of the UK and to phenomena beyond the pandemic.

Product Line Re‐Engineering for Modularity in a US Department of Defense Project

ABSTRACTThis case study details the options evaluated and path chosen by a United States (US) Department of Defense (DoD) software development organization to re‐engineer four existing products with common features into a single product‐line resulting in product sponsors taking advantage of cost savings, developers shortening implementation and testing timeframes, and users obtaining product features faster while sharing a common experience across product variants. Although the software‐intensive products were originally a product line operating from a common code repository, they diverged due to different product sponsors having differing priorities and schedule commitments. The mission scope of each variant differs leading to a commonality range of approximately 20% to 70% based on the quantity of common features. The re‐engineering options evaluated included merging common code and maintaining it in a single repository; re‐using software code while keeping it in separate repositories for each product variant; and pursuing a Modular Open Systems Approach (MOSA) to create common modules for insertion, updating, and replacement within any product variant without disrupting the rest of that product. With product sponsor support, the DoD project decided to pursue a hybrid approach of immediate code re‐use complemented with an agile approach to MOSA implementation. This solution allowed the project to re‐engineer the four existing product variants while still meeting sponsor, DoD, and end‐user operational needs.

Blockchain is dead! Long live Blockchain!

A decade on since Satoshi’s Bitcoin paper, Blockchain is now considered to be sliding into the trough of Gartner’s hype cycle. Claims in regards to Blockchain and Cryptocurrencies being dead are on the rise, whilst at the same time many claim the contrary. The vague statement encapsulates many different aspects and perspectives of a myriad of use cases, technology and platforms including both the technique as a whole as well as individual instantiations. In this paper, we unpack the statement, break it down, and investigate objectively concrete factors which provide indication in regards to whether Blockchain is dead. We examine metrics including budgets and investment; public company registries and data; community engagement, projects, and source code repositories; academic research and programmes; social media posts; and public interest. We demonstrate metrics individually that indicate the respective measures’ healthy activity and come to the conclusion that the collective statement “Blockchain is dead” does not hold. A clear message extracted from the work proposed herein is that success is achieved where the community comes together rather than works in isolation.

MetaPrism: A versatile toolkit for joint taxa/gene analysis of metagenomic sequencing data.

In microbiome research, metagenomic sequencing generates enormous amounts of data. These data are typically classified into taxa for taxonomy analysis, or into genes for functional analysis. However, a joint analysis where the reads are classified into taxa-specific genes is often overlooked. To enable the analysis of this biologically meaningful feature, we developed a novel bioinformatic toolkit, MetaPrism, which can analyze sequence reads for a set of joint taxa/gene analyses to: 1) classify sequence reads and estimate the abundances for taxa-specific genes; 2) tabularize and visualize taxa-specific gene abundances; 3) compare the abundances between groups; and 4) build prediction models for clinical outcome. We illustrated these functions using a published microbiome metagenomics dataset from patients treated with immune checkpoint inhibitor therapy and showed the joint features can serve as potential biomarkers to predict therapeutic responses. MetaPrism is a toolkit for joint taxa and gene analysis. It offers biological insights on the taxa-specific genes on top of the taxa-alone or gene-alone analysis.MetaPrism is open-source software and freely available at https://github.com/jiwoongbio/MetaPrism. The example script to reproduce the manuscript is also provided in the above code repository.

An Empirical Study of Students’ Perceptions on the Setup and Grading of Group Programming Assignments

Courses in computer science curricula often involve group programming assignments. Instructors are required to take several decisions on assignment setup and monitoring, team formation policies, and grading systems. Group programming projects provide unique monitoring opportunities due to the availability of both product and process data, as well as challenges in team composition, with students of varying levels of prior programming experience. To gain insights into the experiences and perceptions of students about the assignment setup and grading policies in group programming projects, we interviewed 20 computer science students from four universities. The thematic analysis highlighted factors in group composition that are considered important, as well as advantages and disadvantages of the self-selection of the teams. It also indicated three grading strategies experienced by the students, namely, being assigned the same group grade, individual grades distributed by the instructor, and grade distribution determined by the team, with perceptions about them varying greatly. Several practices for monitoring team contributions were identified. Checking the source code repositories was considered useful in recognizing slacking members, but automated metrics are not always representative of the work distribution. The analysis also uncovered student perceptions on the grading factors for programming assignments, including coding efficiency and skill.

On the generalizability of Neural Program Models with respect to semantic-preserving program transformations

With the prevalence of publicly available source code repositories to train deep neural network models, neural program models can do well in source code analysis tasks such as predicting method names in given programs that cannot be easily done by traditional program analysis techniques. Although such neural program models have been tested on various existing datasets, the extent to which they generalize to unforeseen source code is largely unknown. Since it is very challenging to test neural program models on all unforeseen programs, in this paper, we propose to evaluate the generalizability of neural program models with respect to semantic-preserving transformations: a generalizable neural program model should perform equally well on programs that are of the same semantics but of different lexical appearances and syntactical structures. We compare the results of various neural program models for the method name prediction task on programs before and after automated semantic-preserving transformations. We use three Java datasets of different sizes and three state-of-the-art neural network models for code, namely code2vec, code2seq, and GGNN, to build nine such neural program models for evaluation. Our results show that even with small semantically preserving changes to the programs, these neural program models often fail to generalize their performance. Our results also suggest that neural program models based on data and control dependencies in programs generalize better than neural program models based only on abstract syntax trees (ASTs). On the positive side, we observe that as the size of the training dataset grows and diversifies the generalizability of correct predictions produced by the neural program models can be improved too. Our results on the generalizability of neural program models provide insights to measure their limitations and provide a stepping stone for their improvement.

A Beginner's Guide to Conducting Reproducible Research

A Beginner's Guide to Conducting Reproducible Research

Code and data repository for Anion exchange capacity explains deep soil nitrate accumulation in Brazilian Amazon croplands

Code and data repository for Anion exchange capacity explains deep soil nitrate accumulation in Brazilian Amazon croplands

Code and data repository for Anion exchange capacity explains deep soil nitrate accumulation in Brazilian Amazon croplands

Code and data repository for Anion exchange capacity explains deep soil nitrate accumulation in Brazilian Amazon croplands

Code and data repository for Anion exchange capacity explains deep soil nitrate accumulation in Brazilian Amazon croplands

Code and data repository for Anion exchange capacity explains deep soil nitrate accumulation in Brazilian Amazon croplands

A Systematic Mapping Study on Analysis of Code Repositories

Code repositories contain valuable information, which can be extracted, processed and synthesized into valuable information. It enabled developers to improve maintenance, increase code quality and understand software evolution, among other insights. Certain research has been made during the last years in this field. This paper presents a systematic mapping study to find, evaluate and investigate the mechanisms, methods and techniques used for the analysis of information from code repositories that allow the understanding of the evolution of software. Through this mapping study, we have identified the main information used as input for the analysis of code repositories (commit data and source code), as well as the most common methods and techniques of analysis (empirical/experimental and automatic). We believe the conducted research is useful for developers working on software development projects and seeking to improve maintenance and understand the evolution of software through the use and analysis of code repositories.

Prediction of GPI-anchored proteins with pointer neural networks

• Deep learning approach for glycosylphosphatidylinositol (GPI) anchoring signal prediction. • A novel, carefully homology partitioned, dataset. • Recurrent neural networks with an attention mechanism. • Exploring biological features uncovered by deep learning. GPI-anchors constitute a very important post-translational modification, linking many proteins to the outer face of the plasma membrane in eukaryotic cells. Since experimental validation of GPI-anchoring signals is slow and costly, computational approaches for predicting them from amino acid sequences are needed. However, the most recent GPI predictor is more than a decade old and considerable progress has been made in machine learning since then. We present a new dataset and a novel method, NetGPI, for GPI signal prediction. NetGPI is based on recurrent neural networks, incorporating an attention mechanism that simultaneously detects GPI-anchoring signals and points out the location of their ω -sites. The performance of NetGPI is superior to existing methods with regards to discrimination between GPI-anchored proteins and other secretory proteins and approximate ( ± 1 position) placement of the ω -site. NetGPI is available at: https://services.healthtech.dtu.dk/service.php?NetGPI . The code repository is available at: https://github.com/mhgislason/netgpi-1.1 .

Software Deterioration Control Based on Issue Reports

Introduction: Successive code changes during the maintenance phase may cause the emergence of bad smells and anti-patterns in code and gradually results in deterioration of the code and difficulties in its maintainability. Continuous Quality Control (QC) is essential in this phase to refactor the anti-patterns and bad smells. Objectives: The objective of this research has been to present a novel component called Code Deterioration Watch (CDW) to be integrated with existing Issue Tracking Systems (ITS) in order to assist the QC team in locating the software modules most vulnerable to deterioration swiftly. The important point regarding the CDW is the fact that its function has to be independent of the code level metrics rather it is totally based on issue level metrics measured from ITS repositories. Methods: An issue level metric that properly alerts us of bad-smell emergence was identified by mining software repositories. To measure that metric, a Stream Clustering algorithm called ReportChainer was proposed to spot Relatively Long Chains (RLC) of incoming issue reports as they tell the QC team that a concentrated point of successive changes has emerged in the software. Results: The contribution of this paper is partly creating a huge integrated code and issue repository of twelve medium and large size open-source software products from Apache and Eclipse. By mining this repository it was observed that there is a strong direct correlation (0.73 on average) between the number of issues of type “New Feature” reported on a software package and the number of bad-smells of types “design” and “error prone” emerged in that package. Besides a strong direct correlation (0.97 on average) was observed between the length of a chain and the magnitude of times it caused changes to a software package. Conclusion: The existence of direct correlation between the number of issues of type “New Feature” reported on a software package and (1) the number of bad-smells of types “design” and “error prone” and (2) the value of “CyclomaticComplexity” metric of the package, justifies the idea of Quality Control merely based on issue-level metrics. A stream clustering algorithm can be effectively applied to alert the emergence of a deteriorated module.

Software development with feature toggles: practices used by practitioners

Background: Using feature toggles is a technique that allows developers to either turn a feature on or off with a variable in a conditional statement. Feature toggles are increasingly used by software companies to facilitate continuous integration and continuous delivery. However, using feature toggles inappropriately may cause problems which can have a severe impact, such as code complexity, dead code, and system failure. For example, the erroneous repurposing of an old feature toggle caused Knight Capital Group, an American global financial services firm, to go bankrupt due to the implications of the resultant incorrect system behavior. Aim: The goal of this research project is to aid software practitioners in the use of practices to support software development with feature toggles through an empirical study of feature toggle practice usage by practitioners. Method: We conducted a qualitative analysis of 99 artifacts from the grey literature and 10 peer-reviewed papers about feature toggles. We conducted a survey of practitioners from 38 companies. Results: We identified 17 practices in 4 categories: Management practices, Initialization practices, Implementation practices, and Clean-up practices. We observed that all of the survey respondents use a dedicated tool to create and manage feature toggles in their code. Documenting feature toggle's metadata, setting up the default value for feature toggles, and logging the changes made on feature toggles are also frequently-observed practices. Conclusions: The feature toggle development practices discovered and enumerated in this work can help practitioners more effectively use feature toggles. This work can enable future mining of code repositories to automatically identify feature toggle practices.

Sketching an AI Marketplace: Tech, Economic, and Regulatory Aspects

Artificial intelligence shows promise for solving many practical societal problems in areas such as healthcare and transportation. However, the current mechanisms for AI model diffusion such as Github code repositories, academic project webpages, and commercial AI marketplaces have some limitations; for example, a lack of monetization methods, model traceability, and model auditabilty. In this work, we sketch guidelines for a new AI diffusion method based on a decentralized online marketplace. We consider the technical, economic, and regulatory aspects of such a marketplace including a discussion of solutions for problems in these areas. Finally, we include a comparative analysis of several current AI marketplaces that are already available or in development. We find that most of these marketplaces are centralized commercial marketplaces with relatively few models.

Extraction of Family History Information From Clinical Notes: Deep Learning and Heuristics Approach

BackgroundElectronic health records store large amounts of patient clinical data. Despite efforts to structure patient data, clinical notes containing rich patient information remain stored as free text, greatly limiting its exploitation. This includes family history, which is highly relevant for applications such as diagnosis and prognosis.ObjectiveThis study aims to develop automatic strategies for annotating family history information in clinical notes, focusing not only on the extraction of relevant entities such as family members and disease mentions but also on the extraction of relations between the identified entities.MethodsThis study extends a previous contribution for the 2019 track on family history extraction from national natural language processing clinical challenges by improving a previously developed rule-based engine, using deep learning (DL) approaches for the extraction of entities from clinical notes, and combining both approaches in a hybrid end-to-end system capable of successfully extracting family member and observation entities and the relations between those entities. Furthermore, this study analyzes the impact of factors such as the use of external resources and different types of embeddings in the performance of DL models.ResultsThe approaches developed were evaluated in a first task regarding entity extraction and in a second task concerning relation extraction. The proposed DL approach improved observation extraction, obtaining F1 scores of 0.8688 and 0.7907 in the training and test sets, respectively. However, DL approaches have limitations in the extraction of family members. The rule-based engine was adjusted to have higher generalizing capability and achieved family member extraction F1 scores of 0.8823 and 0.8092 in the training and test sets, respectively. The resulting hybrid system obtained F1 scores of 0.8743 and 0.7979 in the training and test sets, respectively. For the second task, the original evaluator was adjusted to perform a more exact evaluation than the original one, and the hybrid system obtained F1 scores of 0.6480 and 0.5082 in the training and test sets, respectively.ConclusionsWe evaluated the impact of several factors on the performance of DL models, and we present an end-to-end system for extracting family history information from clinical notes, which can help in the structuring and reuse of this type of information. The final hybrid solution is provided in a publicly available code repository.

Behaviour Driven Testing as an Effective Approach for Automated Testing in Continuous Integration

The global digitalization has transformed business dramatically, making it respond and adapt fast to the changing world and new challenges. The development teams face a global challenge of how to speed up product delivery time to the end user without losing product quality. Due to this, the demand for automation of routine tasks has significantly increased, because this process helps to minimize and optimize time and money for digital products delivery.Functional automation testing plays a very important role in the process of faster and more effective product delivery to the end users in a limited time. For example, test automation plays a key role when you need to test periodically the same functionality over and over again to check whether the product is ready for delivery to the end user or not, especially, when working with very complex systems.The development of the automated testing system architecture is influenced by such factors as the analysis of the possibility of coverage by automated tests, the creation of a strategy to build the automation testing on the project and the development of the automated testing system at its own. The test approach itself also plays an important role and could vary dramatically. For instance, Test-Driven Development, Behaviour-Driven Development, Keyword-Driven Testing, and Data-Driven Testing approaches.The behavioural approach (BDD) has been used to develop an automation testing system described in this work. This approach is a set of practices aimed to reduce the risks with communication between the development team and the client. There is a wide range of tools for BDD-based testing, namely Cucumber, JBehave, Gauge, and more.As a result, the automated testing system, described in this paper, has been built using Cucumber practices, which in turn allows functional verifications to be performed simultaneously at such levels as graphical interface (UI) and web service (testing API). Cucumber recognizes the specification for the tested functionality, which is described in a simple language, then looks whether testing scenarios are present in the feature files and only then runs test scripts. Each scenario consists of a set of steps that Cucumber processes and verifies against a given expected business logic.It should also be noted that this system has been integrated into Continuous Integration, based on the Jenkins service. Test automation framework integration with CI tool helps to constantly monitor the project source code repository for changes, and this enables fast feedback and high quality of the product in the process of continuous development.Manuscript received 04.06.2020