Pseudogen

Pseudogen generates English pseudocode descriptions from Python source code. It supports local Docker builds and manual installation on Ubuntu 24.04. Try the Pseudogen demo.

Install with Docker

Use Docker to build the current checkout, train the model, and run inference:

1. Install Docker.

2. Clone this repository:

   git clone https://github.com/delihiros/pseudogen.git
   cd pseudogen

3. Build the image:

   docker build -t pseudogen .

   The build compiles the tools, downloads and verifies the corpus, and trains the model.

4. Start interactive inference:

   docker run --rm -it pseudogen

   Enter Python source code, then press Ctrl-D to end input.

Run a non-interactive inference command when you already have the input:

printf 'x += 1\n' | docker run --rm -i pseudogen

Install manually on Ubuntu 24.04

Install the build and runtime dependencies on Ubuntu 24.04:

sudo apt-get update
sudo apt-get install -y --no-install-recommends \
  autoconf automake autotools-dev build-essential ca-certificates cmake git \
  libboost-all-dev libtool python3 python3-nltk wget zlib1g-dev

Clone the repository and compile the bundled tools:

git clone https://github.com/delihiros/pseudogen.git
cd pseudogen
./tool_setup.sh

Download the training corpus

Create the data directory, download the annotated Django corpus, and verify its SHA-256 checksum:

mkdir data
cd data
wget -O en-django.tar.gz http://ahclab.naist.jp/pseudogen/en-django.tar.gz
echo 'd86f8e5dcc3411658bb50bf71a8ff0d9673e783769f9f46d99926fb3b515b7a6  en-django.tar.gz' | sha256sum -c -
tar -xzf en-django.tar.gz
mv en-django/all.* .

Train the model

Run training from the data directory:

../train-pseudogen.sh -p all.code -e all.anno

Training creates the model configuration at tune/travatar.ini.

Evaluate the model

Run evaluation from the data directory:

../tools/travatar/src/bin/travatar -threads 2 -config_file tune/travatar.ini < test.reducedtree > test.hyp
../test-pseudogen.sh -r test.entok -h test.hyp

The first command generates translations. The second reports BLEU and RIBES scores.

Run inference manually

Run a single inference from the data directory:

printf 'x += 1\n' | ../run-pseudogen.sh -f tune/travatar.ini

You can also run ../run-pseudogen.sh -f tune/travatar.ini interactively. Enter Python source code, then press Ctrl-D to end input.

How Pseudogen works

Pseudogen aligns Python and English training data, then trains a tree-to-string machine translation model.

Papers

• Methods, IEEE/ACM ASE 2015
• Software, IEEE/ACM ASE 2015

Tools used

• GIZA++ creates alignments
• Travatar trains the tree-to-string machine translation model
• mteval evaluates translations

Contributors

• Yusuke Oda
• Hiroyuki Fudaba
• Koichi Akabe
• Graham Neubig
• Hideaki Hata