Automatic Generation of Text Descriptive Comments for Code Blocks

Authors: Yuding Liang, Kenny Zhu

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our evaluation comes in two parts. In the first part, we evaluate Code-RNN model s ability to classify different source code blocks into k known categories. In the second part, we show the effectiveness of our comment generation model by comparing with several state-of-the-art approaches in both quantitative and qualitative assessments. the overall framework achieves remarkable accuracy (Rouge-2 value) in the task of generating descriptive comments for Java methods, compared to state-of-the-art approaches.
Researcher Affiliation Academia Yuding Liang, Kenny Q. Zhu liangyuding@sjtu.edu.cn, kzhu@cs.sjtu.edu.cn Department of Computer Science and Engineering Shanghai Jiao Tong University 800 Dongchuan Road, Shanghai, China 200240
Pseudocode No The paper includes equations and diagrams describing the models (Code-RNN, Code-GRU) but does not contain explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes The source code of our approach as well as all data set is available at https://adapt.seiee.sjtu.edu.cn/CodeComment/.
Open Datasets Yes The source code of our approach as well as all data set is available at https://adapt.seiee.sjtu.edu.cn/CodeComment/. Our data set comes from the Google Code Jam contest (2008 – 2016), which there are multiple problems... We use the solutions (10,724 methods) of 6 problems as training set and the solutions (30 methods) of the other 6 problems as the test set. For one project, we separate the commented methods into three parts: training set, validation set and test set.
Dataset Splits Yes For one project, we separate the commented methods into three parts: training set, validation set and test set. We tune these two parameters on the validation set to determine which values to use.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions tools like 'Java Parser' and discusses baselines like 'Moses' and 'CODE-NN', but does not provide specific version numbers for the key software dependencies or libraries (e.g., Python, deep learning frameworks like TensorFlow/PyTorch) used to replicate the experiments.
Experiment Setup Yes Seq2Seq, Basic RNN and our model run 800 epochs during training time. We tune these two parameters on the validation set to determine which values to use. Our tuning ranges are: beam size: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10] weight for the length penalty: [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]