Code for the engine implementation and reproducibility of the evaluation in paper (LINK). Allows the dynamic loading of just-in-time compiled Python UDF code into a running vectorized query engine.
Just use the provided Dockerfile to generate the environment for the experiments.
docker build -t compiledudfs .
docker run -it compiledudfs
In the root folder of the running docker container, run
BUILDMODE=RELEASE . build.sh
to build the engine and in release mode and set the environment.
Run tests using
cd /release
make test
or run the experiments using
cd /exp
python3.6 run_experiments.py <number_of_runs>
which prints the dataframe normalized against python execution time, writes all measurements to /tmp/results.csv and saves the generated plot from the paper in /tmp/results.png.