DeepClone: Modeling Clones to Generate Code Predictions

Abstract

During so ware development, programmers o en tend to reuse the code for common functionalities, available in other source code repositories. is activity helps them to reduce time and e ort to develop code, instead of building it from scratch. Code clones are candidates for reuse in an exploratory or rapid development, as they represent o en repeated functionality in so ware systems. To facilitate code clone reuse, we propose a novel approach, Deep-Clone, where we utilize a deep learning algorithm for modeling code clones and predicting the next possible set of tokens (up to the cloned method body) based on the user input so far. e generated predictions aim to potentially help developers to write code rapidly with minimum tuning of values later on. DeepClone applies natural language processing techniques to learn from a large code corpus (the BigCloneBench dataset), and generates code tokens (full clone methods where applicable) using the model learned. We have quantitatively evaluated our solution to assess (1) our model's quality and its accuracy in token prediction, and (2) its performance and e ectiveness in clone method prediction. With a high quality and accurate model as the foundation, we further discuss scenarios for exploiting our approach. CCS CONCEPTS•Computing methodologies → Neural networks; •So ware and its engineering → Language features; Reusability; Automatic programming; So ware development methods;

Review: Deep Learning in Electron Microscopy

“…Similar DNNs can assist programmers 1609,1610 . For example, by predictive source code completion [1611][1612][1613][1614][1615][1616] , and by generating source code to map inputs to target outputs 1617 or from labels describing desired source code 1618 . Text generating DNNs can also help write scientific papers.…”

Augmenting Machine Learning with Information Retrieval to Recommend Real Cloned Code Methods for Code Completion

“…There are slight variations in the perplexity values of Top-10 samples, which can be attributed to various factors such as the type of functionality, the number of clone method snippet trained in the DeepClone model, and inner similarity among the clone methods' type. These factors have been discussed in detail in our previous work [17].…”

“…We have also developed a search corpus comprising of real clone methods from those datasets. Initially, independent of the search corpus, we attempt to predict clone methods consisting of subsequent token sequences starting from the start-of-clone tag soc until the end-of-clone tag eoc ( [17]). Then, we apply an IR technique to retrieve real clone methods from the search corpus, which are most similar to the initially generated DeepClone prediction.…”

“…Then, we apply an IR technique to retrieve real clone methods from the search corpus, which are most similar to the initially generated DeepClone prediction. Figure 1 displays a pictorial representation of our methodology to generate real clone methods (without showing the training steps of DeepClone model, which are shown in our previous paper [17]). In their raw form, we obtain DeepClone output, ground truth, and top-10 samples from the current methodology in an un-formatted style, where code tokens of a clone method are separated only by space characters.…”

“…We have performed several pre-processing steps to build our search corpus, which is similar to what we have followed in our previous work [17]. First, we have extracted the details of a total of 14,922 true positive clone methods, in which 11,920 are distinct clone methods (Extraction).…”