Awesome Machine Learning On Source Code

A curated list of awesome machine learning frameworks and algorithms that work on top of source code. Inspired by Awesome Machine Learning.

Contents

• Digests
• Articles
  • Papers
  • Posts
  • Talks
• Software
  • Machine Learning
  • Utilities
• Datasets
• Credits
• Contributions
• License

Digests

• Learning from "Big Code"
• A Survey of Machine Learning for Big Code and Naturalness

Articles

Papers

• Topic modeling of public repositories at scale using names in source code
• Parameter-Free Probabilistic API Mining across GitHub
• A Subsequence Interleaving Model for Sequential Pattern Mining
• A Convolutional Attention Network for Extreme Summarization of Source Code
• Parameter-Free Probabilistic API Mining across GitHub
• Tailored Mutants Fit Bugs Better
• TASSAL: Autofolding for Source Code Summarization
• Suggesting Accurate Method and Class Names
• Mining idioms from source code
• Mining Source Code Repositories at Massive Scale using Language Modeling
• Why, When, and What: Analyzing Stack Overflow Questions by Topic, Type, and Code
• Latent Predictor Networks for Code Generation - Addresses the problem of generating programming code from a mixed natural language and structured specification. Wang Ling, Edward Grefenstette, Karl Moritz Hermann, Tomáš Kočiský, Andrew Senior, Fumin Wang, Phil Blunsom.
• Code Completion with Statistical Language Models - Veselin Raychev, Martin Vechev, Eran Yahav.
• Using recurrent neural networks to predict next tokens in the java solutions - Alex Skidanov, Illia Polosukhin.
• Learning Python Code Suggestion with a Sparse Pointer Network - Avishkar Bhoopchand, Tim Rocktäschel, Earl Barr, Sebastian Riedel.
• Learning Efficient Algorithms with Hierarchical Attentive Memory - Andrychowicz, Marcin, and Karol Kurach.
• DeepCoder: Learning to Write Programs - Balog, Matej, Alexander L. Gaunt, Marc Brockschmidt, Sebastian Nowozin, and Daniel Tarlow.
• Programming with a Differentiable Forth Interpreter - Bošnjak, Matko, Tim Rocktäschel, Jason Naradowsky, and Sebastian Riedel.
• Learning to Superoptimize Programs - Workshop Version - Bunel, Rudy, Alban Desmaison, M. Pawan Kumar, Philip H. S. Torr, and Pushmeet Kohli.
• Meta-Interpretive Learning of Efficient Logic Programs - Cropper, Andrew, and Stephen H. Muggleton.
• Learning Operations on a Stack with Neural Turing Machines - Deleu, Tristan, and Joseph Dureau.
• Neural Functional Programming - Feser, John K., Marc Brockschmidt, Alexander L. Gaunt, and Daniel Tarlow.
• TerpreT: A Probabilistic Programming Language for Program Induction - Gaunt, Alexander L., Marc Brockschmidt, Rishabh Singh, Nate Kushman, Pushmeet Kohli, Jonathan Taylor, and Daniel Tarlow.
• Neural Turing Machines - Graves, Alex, Greg Wayne, and Ivo Danihelka.
• Neural Symbolic Machines: Learning Semantic Parsers on Freebase with Weak Supervision (Short Version) - Liang, Chen, Jonathan Berant, Quoc Le, Kenneth D. Forbus, and Ni Lao.
• Probabilistic Neural Programs - Murray, Kenton W., and Jayant Krishnamurthy.
• Neural Programmer: Inducing Latent Programs with Gradient Descent - Neelakantan, Arvind, Quoc V. Le, and Ilya Sutskever.
• Divide and Conquer with Neural Networks - Nowak, Alex, and Joan Bruna.
• Neural Programmer-Interpreters - Reed, Scott, and Nando de Freitas.
• Programs as Black-Box Explanations - Singh, Sameer, Marco Tulio Ribeiro, and Carlos Guestrin.
• A Differentiable Approach to Inductive Logic Programming - Yang, Fan, Zhilin Yang, and William W. Cohen.
• From Machine Learning to Machine Reasoning - Bottou, Leon.
• Learning Latent Multiscale Structure Using Recurrent Neural Networks - Chung, Junyoung, Sungjin Ahn, and Yoshua Bengio.
• Lifelong Perceptual Programming By Example - Gaunt, Alexander L., Marc Brockschmidt, Nate Kushman, and Daniel Tarlow.
• Inferring Algorithmic Patterns with Stack-Augmented Recurrent Nets - Joulin, Armand, and Tomas Mikolov.
• Neural GPUs Learn Algorithms - Kaiser, Łukasz, and Ilya Sutskever.
• API usage pattern recommendation for software development - Haoran Niu, Iman Keivanloo, Ying Zou.
• Summarizing Source Code using a Neural Attention Model University of Washington CSE, Seatle, WA, USA.
• Program Synthesis from Natural Language Using Recurrent Neural Networks - University of Washington CSE, Seatle, WA, USA.
• Exploring API Embedding for API Usages and Applications - Nguyen, Nguyen, Phan and Nguyen.
• Neural Nets Can Learn Function Type Signatures From Binaries - Zheng Leong Chua, Shiqi Shen, Prateek Saxena, and Zhenkai Liang.
• Deep Learning Code Fragments for Code Clone Detection - Martin White, Michele Tufano, Christopher Vendome, and Denys Poshyvanyk.
• Automated Identification of Security Issues from Commit Messages and Bug Reports [PDF] Yaqin Zhou and Asankhaya Sharma.

Posts

• Weighted MinHash on GPU helps to find duplicate GitHub repositories.
• Source Code Identifier Embeddings
• The half-life of code & the ship of Theseus
• The eigenvector of "Why we moved from language X to language Y"
• Analyzing Github, How Developers Change Programming Languages Over Time

Talks

• Topic Modeling of GitHub Repositories
• Similarity of GitHub Repositories by Source Code Identifiers
• Using deep RNN to model source code
• Source code abstracts classification using CNN (1)
• Source code abstracts classification using CNN (2)
• Source code abstracts classification using CNN (3)
• Embedding the GitHub contribution graph

Software

Machine Learning

• Differentiable Neural Computer (DNC) - TensorFlow implementation of the Differentiable Neural Computer.
• sourced.ml - Abstracts feature extraction from source code syntax trees and working with ML models.
• vecino - Finds similar Git repositories.
• apollo - Source code deduplication as scale, research.
• gemini - Source code deduplication as scale, production.
• enry - Insanely fast file based programming language detector.
• Naturalize - Language agnostic framework for learning coding conventions from a codebase and then expoiting this information for suggesting better identifier names and formatting changes in the code.
• Extreme Source Code Summarization - Convolutional attention neural network that learns to summarize source code into a short method name-like summary by just looking at the source code tokens.
• Summarizing Source Code using a Neural Attention Model - CODE-NN, uses LSTM networks with attention to produce sentences that describe C# code snippets and SQL queries from StackOverflow. Torch over C#/SQL
• Probabilistic API Miner - Near parameter-free probabilistic algorithm for mining the most interesting API patterns from a list of API call sequences.
• Interesting Sequence Miner - Novel algorithm that mines the most interesting sequences under a probabilistic model. It is able to efficiently infer interesting sequences directly from the database.
• TASSAL - Tool for the automatic summarization of source code using autofolding. Autofolding automatically creates a summary of a source code file by folding non-essential code and comment blocks.
• JNice2Predict - Efficient and scalable open-source framework for structured prediction, enabling one to build new statistical engines more quickly.

Utilities

• go-git - Highly extensible Git implementation in pure Go which is friendly to data mining.
• hercules - Git repository mining framework with batteries on top of go-git.
• bblfsh - Self-hosted server for source code parsing.
• engine - Scalable and distributed data retrieval pipeline for source code.
• minhashcuda - Weighted MinHash implementation on CUDA to efficiently find duplicates.
• kmcuda - k-means on CUDA to cluster and to search for nearest neighbors in dense space.
• wmd-relax - Python package which finds nearest neighbors at Word Mover's Distance.

Datasets

• GitHub repositories - languages distribution - Programming languages distribution in 14,000,000 repositories on GitHub (October 2016).
• 452M commits on GitHub - ≈ 452M commits' metadata from 16M repositories on GitHub (October 2016).
• GitHub readme files - Readme files of all GitHub repositories (16M) (October 2016).
• from language X to Y - Cache file Erik Bernhardsson collected for his awesome blog post.
• GitHub word2vec 120k - Sequences of identifiers extracted from top starred 120,000 GitHub repos.
• GitHub Source Code Names - Names in source code extracted from 13M GitHub repositories, not people.
• GitHub duplicate repositories - GitHub repositories not marked as forks but very similar to each other.
• GitHub lng keyword frequencies - Programming language keyword frequency extracted from 16M GitHub repositories.
• GitHub Java Corpus - GitHub Java corpus is a set of Java projects collected from GitHub that we have used in a number of our publications. The corpus consists of 14,785 projects and 352,312,696 LOC.
• 150k Python Dataset - Dataset consisting of 150'000 Python ASTs.
• 150k JavaScript Dataset - Dataset consisting of 150'000 JavaScript files and their parsed ASTs.
• card2code - This dataset contains the language to code datasets described in the paper Latent Predictor Networks for Code Generation.

Credits

• A lot of references and articles were taken from mast-group

Contributions

See CONTRIBUTING.md.

License