Enso Dataframes Benchmarks

A set of benchmarks that compares Enso's Table library performance against analogous operations in Python's Pandas.

The benchmarks are implemented in src/Main.enso and python/baseline.py respectively. Related operations have analogous function names in both languages and they were implemented using analogous constructs in both languages.

Running the Benchmarks

To run the benchmarks you need:

• Python 3.7+ installed and available on your PATH as python.
• pandas version 1.1.5,
• Enso engine and launcher, version 0.2.0, built from commit d2df28bc43fcfee614fb31780cd46a59ebb16127, with launcher enso executable on your PATH,
• wget on your PATH if you want the script to download input data automatically.

You can use pyenv to manage python versions:

> pyenv install 3.7.9

If you want to use this Python version for processing the results in Jupyter Notebook as described later, make sure that sqlite is properly set up. For example, on Ubuntu you may need to apt install libsqlite3-dev before installing Python.

To install pandas and additional dependencies that are useful for generating the report, you can use poetry:

# Go to the root directory of the repository and type:
> poetry install
> poetry shell # to enter the shell that uses poetry's virtual environment

You can download just the launcher or the whole bundle for 0.2.0 from the enso repository. After extracting it, you need to add the launcher executable to your PATH. If you downloaded just the launcher, make sure that the engine is installed by running enso install engine 0.2.0.

After ensuring that all the dependencies are installed, you can run the benchmarks with:

> ./run_benchmarks.py

If you are using poetry, make sure that you run the above command inside of poetry shell or otherwise use:

poetry run ./run_benchmarks.py

The script will automatically download all input data and run the benchmarks, saving results to report/ directory. If you don't have wget, you can manually download the input files as described in ./data/README.md.

Processing the Results

The ./run_benchmarks.py will generate files enso_times.csv and python_times.csv that contain time measurements for each tested operation. To generate plots comparing these measurements you can use a prepared Jupyter Notebook at ./report/Report.ipynb. It will load the data and create the relevant graphs.

To use it, you need jupyter installed - it will be available if you have used poetry for resolving benchmark dependencies as described above.

To prepare the report, first start Jupyter Notebook with:

> cd report/
> poetry run jupyter notebook

Then you can open the notebook Report.ipynb and select Kernel > Restart and Run All to generate the report based on latest measurements. The generated report can be exported with File > Download As.