Extra notes on installation:

• Torch installation:
  • to overcome the LuaJIT low memory limit when using torch.save, install torch with Lua 5.2 instead of LuaJIT
    use ./clean.sh; TORCH_NVCC_FLAGS="-D__CUDA_NO_HALF_OPERATORS__" CC=/usr/bin/gcc-6 CXX=/usr/bin/g++-6 TORCH_LUA_VERSION=LUA52 ./install.sh to install Torch.
  • (build-essential, gcc, g++, libreadline-dev, cmake)
• install cuda
• Cutorch installation: torch/cutorch
  luarocks install cutorch, luarocks install cunn
• (for data preprocessing) antlr4: we need version 4.5 specified in src/csharp/CSharp4Lexer.py
  python2 -m pip install 'antlr4-python2-runtime>=4.5,<4.6'
• (for data preprocessing) antlr4-complete: to create a parser for Java, we need to download:
  curl -O https://www.antlr.org/download/antlr-4.5.3-complete.jar
export CLASSPATH=".:[path: antlr-4.5.3-complete.jar]:$CLASSPATH"

Run CODENN

See details of CODENN in our paper

Summarizing Source Code using a Neural Attention Model (https://github.com/sriniiyer/codenn/blob/master/summarizing_source_code.pdf)

Requirements

• Torch (https://torch.ch/docs/getting-started.html)
• Cutorch
• antlr4 for parsing C# (pip install antlr4-python2-runtime)

Setup environment

export PYTHONPATH=~/codenn/src/:~/codenn/src/sqlparse export CODENN_DIR=~/codenn/ export CODENN_WORK=./workdir

Build both csharp and sql datasets

Install modified sqlparse

cd src/sqlparse/ sudo python setup.py install

Build datasets

cd src/model ./buildData.sh

Train codenn models and predict on test set

./run.sh {sql|csharp}