Omniocular is a framework for building deep learning models on code, implemented in PyTorch by the Data Systems Group at the University of Waterloo. Various modules in Omniocular are heavily inspired by (and are compatible with) Hedwig, a framework for dcoument classification.

Models

Predictions over a single sequence of tokens

• Reg-CNN: Convolutional networks with regularization
• Reg-LSTM: Regularized LSTM for token sequence classification
• HR-CNN: Hierarchical Convolutional Networks with regularization

Predictions over a paired sequence of tokens

• Reg-CNN: Convolutional networks with regularization
• Reg-LSTM: Regularized LSTM for token sequence classification
• HR-CNN: Hierarchical Convolutional Networks with regularization

Embeddings for code

• Token2vec: Word2vec-based embeddings for programming language tokens
• Code2vec: Distributed representations for code from collections of AST paths

Each model directory has a README.md with further details.

Setting up PyTorch

Omniocular is designed for Python 3.6 and PyTorch 0.4. PyTorch recommends Anaconda for managing your environment. We recommend creating a custom environment as follows:

$ conda create --name omniocular python=3.6
$ source activate omniocular

And installing PyTorch as follows:

$ conda install pytorch=0.4.1 cuda92 -c pytorch

Other Python packages we use can be installed via pip:

$ pip install -r requirements.txt

Datasets

Download the datasets and embeddings from the omniocular-data repository:

$ git clone https://github.com/omniocular/omniocular.git
$ git clone https://git.uwaterloo.ca/arkeshav/omniocular-data.git

Datasets, along with embeddings should be placed in the omniocular-data folder, with the following directory structure:

.
├── omniocular
└── omniocular-data
    ├── embeddings
    └── datasets

After cloning the omniocular-data repo, check if you have a text file containing the embeddings for Java:

cd omniocular-data/embeddings/
ls java1k_size300_min10.bin.txt