System for synthesizing Java API idioms, powered by Neural Sketch Learning
Latest commit 22d94fd Jul 12, 2018


Bayou is a data-driven program synthesis system for Java API idioms that uses the novel technique of Neural Sketch Learning.

To learn more, click here for the ICLR 2018 paper on Bayou. Click here for the PLDI 2018 tutorial slides on Bayou.

Try Bayou on your browser! Visit


  • JDK 1.8
  • Python3 (Tested with 3.5.1)
  • Tensorflow (Tested with 1.4)
  • scikit-learn (Tested with 0.19)

Compiling and Running Bayou from Source on Ubuntu/Mac

1.) Download source from GitHub:

git clone

2.) Install Dependencies

cd bayou/tool_files/build_scripts
sudo ./

3.) Compile Bayou


4.) Run Bayou

cd out/
./ &

Wait until you see:

            Bayou Ready            

then execute:


You should see as output several programs separated by the characters

---------- BEGIN PROGRAM  ----------

To run Bayou on some of the provided examples, execute:

./ example_inputs/<>

where is the name of the example program.