Testing the performance of Rust and Python on a simple data science task.
Blog Post: Rust vs Python on Data Science, but why not both?
Tools and Requirements:
- https://github.com/dgrunwald/rust-cpython
- https://github.com/PyO3/setuptools-rust
- https://docs.rs/crate/pyo3/0.13.1
- brew install gnu-time
- virtualenv
- Python 3.7.4
Importante Sources:
Running the tests:
$ virtualenv env
$ source venv/bin/activate
$ pip install
Results:
Tests ran on a macOS Catalina - 2.4 GHz 8-Core Intel Core i9 - 32 GB 2667 MHz DDR4
# Python invoking pure Rust
$ gtime python entropy_rust.py
3.00user 0.61system 0:02.01elapsed 180%CPU (0avgtext+0avgdata 60240maxresident)k
0inputs+0outputs (5major+15215minor)pagefaults 0swaps
# Python with Data Science Libraries
$ gtime python entropy_python_data_science.py
5.85user 1.28system 0:05.84elapsed 122%CPU (0avgtext+0avgdata 151744maxresident)k
0inputs+0outputs (2133major+35969minor)pagefaults 0swaps
# Pure Python
$ gtime python entropy_pure_python.py
83.88user 0.95system 1:23.82elapsed 101%CPU (0avgtext+0avgdata 60004maxresident)k
0inputs+0outputs (2129major+13035minor)pagefaults 0swaps
Benchmark Tests:
$ pytest --benchmark-min-rounds=120 --benchmark-histogram=histogram entropy.py
---------------------------------------------------------------------------------------------------- benchmark: 6 tests ----------------------------------------------------------------------------------------------------
Name (time in us) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_rust 542.1070 (1.0) 866.5210 (1.11) 580.0547 (1.02) 43.2468 (1.16) 568.4570 (1.02) 7.0382 (1.0) 121;316 1,723.9754 (0.98) 1333 1
test_pure_numpy_numba 548.9560 (1.01) 783.2350 (1.0) 567.3531 (1.0) 37.2231 (1.0) 554.9335 (1.0) 10.8635 (1.54) 9;13 1,762.5707 (1.0) 120 1
test_python_scipy_numpy 2,832.0430 (5.22) 4,264.0880 (5.44) 3,235.2110 (5.70) 304.5976 (8.18) 3,188.9715 (5.75) 381.0620 (54.14) 34;6 309.0988 (0.18) 146 1
test_python_numpy 2,850.6520 (5.26) 4,423.0050 (5.65) 3,100.2738 (5.46) 197.0626 (5.29) 3,052.7360 (5.50) 190.6520 (27.09) 36;13 322.5522 (0.18) 302 1
test_python_numpy_numexpr 2,967.3900 (5.47) 4,639.2210 (5.92) 3,213.1512 (5.66) 223.0687 (5.99) 3,164.0450 (5.70) 170.1362 (24.17) 18;12 311.2210 (0.18) 229 1
test_pure_python 171,827.4220 (316.96) 204,286.8360 (260.82) 182,864.1219 (322.31) 5,189.4670 (139.42) 181,877.5370 (327.75) 6,761.0355 (960.61) 34;3 5.4685 (0.00) 120 1
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Legend:
Outliers: 1 Standard Deviation from Mean; 1.5 IQR (InterQuartile Range) from 1st Quartile and 3rd Quartile.
OPS: Operations Per Second, computed as 1 / Mean
