Performance Prediction for Lock-based Programs DOI
Dongwen Zhang, Tongtong Wang, Yang Zhang

et al.

Published: Oct. 22, 2023

Using different locking mechanisms affects parallel programs' performance differently, and its impact on program is difficult to assess, which hinders researchers from rationally utilizing mechanisms. Moreover, there are few studies the prediction of To address this issue, paper proposes a combination deep feedforward neural network (FNN) Random Forest (RF) method LockPerf predict programs with The predicted in execution time program. In paper, extracting static characteristics first, then sets variables such as number threads, lock type, read/write ratio by switch statement, finally runs collect multiple samples construct configurable data set. A total 9 projects employed evaluate effectiveness Experimental results show that average mean relative errors 5.47, standard 95% confidence intervals 0.13. experiments effectively predicts

Language: Английский

Code smell detection based on supervised learning models: A survey DOI
Yang Zhang,

Chuyan Ge,

Haiyang Liu

et al.

Neurocomputing, Journal Year: 2023, Volume and Issue: 565, P. 127014 - 127014

Published: Nov. 9, 2023

Language: Английский

Citations

6
Examining deep learning’s capability to spot code smells: a systematic literature review DOI
Ruchika Malhotra, Bhawna Jain,

Marouane Kessentini

et al.

Cluster Computing, Journal Year: 2023, Volume and Issue: 26(6), P. 3473 - 3501

Published: Oct. 3, 2023

Language: Английский

Citations

5
Revisiting "code smell severity classification using machine learning techniques" DOI
Wenhua Hu, Lei Liu, Peixin Yang

et al.

2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC), Journal Year: 2023, Volume and Issue: unknown, P. 840 - 849

Published: June 1, 2023

In the context of limited maintenance resources, predicting severity code smells is more practically useful than simply detecting them. Fontana et al. first empirically investigated some classification algorithms and regression algorithms, for prediction. Their results showed that random forest decision tree performed well on Mean Absolute Error (MAE), Squared (MSE), Spearman Kendall rank correlation coefficients. However, they did not consider issue imbalanced data distribution in dataset, used inappropriate performance evaluation metrics. Therefore, we revisit effectiveness 10 methods 11 methods, prediction using Cumulative Lift Chart (CLC) Severity@20% as primary metrics Accuracy secondary indicator. The show Gradient Boosting Regression (GBR) method performs best terms these

Language: Английский

Citations

4
Application of Deep Learning for Code Smell Detection: Challenges and Opportunities DOI

Mouna Hadj-Kacem,

Nadia Bouassida

SN Computer Science, Journal Year: 2024, Volume and Issue: 5(5)

Published: June 3, 2024

Language: Английский

Citations

1
Automatic detection of code smells using metrics and CodeT5 embeddings: a case study in C# DOI Creative Commons
Aleksandar Kovačević, Nikola Luburić, Јелена Сливка

et al.

Published: May 4, 2022

Code smells are code structures that harm the software’s quality. An obstacle to developing automatic detectors is available datasets' limitations. Furthermore, researchers developed many solutions for Java while neglecting other programming languages. Recently, we created smell dataset C# by following an annotation procedure inspired established practices in Natural Language Processing. This paper evaluates Machine Learning (ML) detection approaches on our novel dataset. We consider two feature representations train ML models: (1) metrics and (2) CodeT5 embeddings. study first state-of-the-art neural source embedding C#. To prove effectiveness of ML, multiple metrics-based heuristics as alternatives. In experiments, best-performing approach was classifier trained (F-measure 0.87 Long Method 0.91 Large Class detection). However, performance improvement over features negligible if advantages automatically inferring features. showed model exceeds human could be helpful developers. best knowledge, this compare against performance.

Language: Английский

Citations

4
On the Relative Value of Imbalanced Learning for Code Smell Detection DOI Open Access
Xiao Yu, Fuyang Li,

Kuan Zou

et al.

Authorea (Authorea), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 10, 2023

Machine learning-based code smell detection has been demonstrated to be a valuable approach for improving software quality and enabling developers identify problematic patterns in code. However, previous researches have shown that the datasets commonly used train these models are heavily imbalanced. While some recent studies explored use of imbalanced learning techniques detection, they only evaluated limited number thus their conclusions about most effective methods may biased inconclusive. To thoroughly evaluate effect on machine we examine 31 with seven classifiers build four data sets. We employ evaluation metrics assess performance Wilcoxon signed-rank test Cliff’s δ . The results show (1) Not all significantly improve performance, but deep forest outperforms other (2) SMOTE (Synthetic Minority Over-sampling TEchnique) is not technique resampling (3) best-performing top-3 little time cost detection. Therefore, provide practical guidelines. First, researchers practitioners should select appropriate (e.g., forest) ameliorate class imbalance problem. In contrast, blind application could harmful. Then, better than selected preprocess

Language: Английский

Citations

2
Code Smell Detection Using a Weighted Cockroach Swarm Optimization Algorithm DOI

G. Saranya,

Dibyajyoti Mishra,

Vojjala Srikar

et al.

2022 13th International Conference on Computing Communication and Networking Technologies (ICCCNT), Journal Year: 2023, Volume and Issue: unknown

Published: July 6, 2023

Code smells are a common problem in software development, leading to decreased maintainability, poor performance, and reduced code quality. Addressing is crucial ensuring that maintainable, scalable, efficient. To detect them, different algorithms used, including evolutionary such as Hybrid Particle Swarm Optimization (HPSOM), Genetic Algorithm (GA), Parallel Evolutionary (PEA), the weighted cockroach swarm optimization algorithm (WCSO). WCSO especially effective for detecting smells, it generates rules combine metrics thresholds automatically. The main focus of this paper using WCSO. We empirically evaluated with other blob, spaghetticode, dataclass, featureenvy, functionaldecomposition. Our results prove outperforms algorithms, GA, HPSOM, PEA. Furthermore, we found performs better on range open-source applications, JfreeChart, GanttProject, Apache Ant 7.0, 5.2, Nutch, Log4J, Lucene, Xerces-J, Rhino. By these can take steps address refactoring make more modular or reducing complexity improve maintainability performance. In conclusion, establishes effectiveness development making reliable

Language: Английский

Citations

2
DeepLog: Deep-Learning-Based Log Recommendation DOI
Yang Zhang,

Xiaosong Chang,

Lining Fang

et al.

2022 IEEE/ACM 44th International Conference on Software Engineering: Companion Proceedings (ICSE-Companion), Journal Year: 2023, Volume and Issue: unknown, P. 88 - 92

Published: May 1, 2023

Log recommendation plays a vital role in analyzing run-time issues including anomaly detection, performance monitoring, and security evaluation. However, existing deeplearning-based approaches for log suffer from insufficient features low F 1 . To this end, paper proposes prototype called DeepLog to recommend location based on deep learning model. parses the source code into an abstract syntax tree then converts each method block hierarchical which extracts both semantic syntactic features. By doing this, we construct dataset with more than 110K samples. employs double-branched neural network model locations. We evaluate effectiveness of by answering four research questions. The experimental results demonstrate that it can 8,725 logs 23 projects is 28.17% higher approaches, improves state-of-the-art.

Language: Английский

Citations

1
ReInstancer: An automatic refactoring approach for Instanceof pattern matching DOI
Yang Zhang,

Shuai Hong

Journal of Software Evolution and Process, Journal Year: 2024, Volume and Issue: 36(8)

Published: March 12, 2024

Abstract The instanceof pattern matching can improve the code quality and readability by removing redundant typecasting simplifying design in different scenarios. However, existing works do not provide sufficient support for refactoring matching. This paper first identifies several cases that cannot be well handled IDEs. Based on these observations, we propose a novel approach called ReInstancer to refactor automatically. conducts program analysis multi‐branch statements obtain variables. After analyzing patterns, are optimized finally refactored into switch or expressions. is evaluated 20 real‐world projects with more than 7,700 experimental results demonstrate total of 3,558 expressions 228 within 10.8 s average each project. improves reducing typecasting, demonstrating its effectiveness.

Language: Английский

Citations

0
Automatic detection of code smells using metrics and CodeT5 embeddings: a case study in C# DOI Creative Commons
Aleksandar Kovačević, Nikola Luburić, Јелена Сливка

et al.

Published: Aug. 9, 2022

Code smells are code structures that harm the software’s quality. An obstacle to developing automatic detectors is available datasets' limitations. Furthermore, researchers developed many solutions for Java while neglecting other programming languages. Recently, we created smell dataset C# by following an annotation procedure inspired established practices in Natural Language Processing. This paper evaluates Machine Learning (ML) detection approaches on our novel dataset. We consider two feature representations train ML models: (1) metrics and (2) CodeT5 embeddings. study first state-of-the-art neural source embedding C#. To prove effectiveness of ML, multiple metrics-based heuristics as alternatives. In experiments, best-performing approach was classifier trained (F-measure 0.87 Long Method 0.91 Large Class detection). However, performance improvement over features negligible if advantages automatically inferring features. showed model exceeds human could be helpful developers. best knowledge, this compare against performance.

Language: Английский

Citations

2