2019/10/05 by Rajas Agashe, Srinivasan Iyer, Agashe, Rajas +3 · 1 citation
Computer Science · #Advanced Malware Detection Techniques #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Software Engineering (cs.SE) #Software Engineering Research #Software Testing and Debugging Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1910.02216
openalex publication_date 2019/10/05 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
Interactive programming with interleaved code snippet cells and natural\nlanguage markdown is recently gaining popularity in the form of Jupyter\nnotebooks, which accelerate prototyping and collaboration. To study code\ngeneration conditioned on a long context history, we present JuICe, a corpus of\n1.5 million examples with a curated test set of 3.7K instances based on online\nprogramming assignments. Compared with existing contextual code generation\ndatasets, JuICe provides refined human-curated data, open-domain code, and an\norder of magnitude more training data. Using JuICe, we train models for two\ntasks: (1) generation of the API call sequence in a code cell, and (2) full\ncode cell generation, both conditioned on the NL-Code history up to a\nparticular code cell. Experiments using current baseline code generation models\nshow that both context and distant supervision aid in generation, and that the\ndataset is challenging for current systems.\n