Skip to content

Benchmarking

This page describes how to benchmark the performance of the grid-fit library against standard Python methods.

How to Run the Benchmark

  1. Make sure you have the required dependencies:
  2. numpy
  3. scipy

You can install them with: 

python -m pip install numpy scipy

  1. Navigate to the examples/benchmarking/ directory.

  2. Run the benchmark script: 

    python benchmark_trilinear.py

This script compares the performance of the standard Python method (scipy.interpolate.RegularGridInterpolator) with the grid-fit trilinear interpolation (to be implemented). The script will print timing results for each method.

Results

Results are a work in progress. The trilinear method is faster than scipy at both construction and interpolation. Derivatives, slices and tricubic to follow...

Benchmark Results Summary

image below shows the benchmark results for various grid sizes and sample counts.

benchmarking results

Grid size:50,   Samples:500     Interp:gridfit is faster by 5.12x       Construct:gridfit is faster by 4.38x
Grid size:100,  Samples:500     Interp:gridfit is faster by 5.15x       Construct:gridfit is faster by 3.16x
Grid size:200,  Samples:500     Interp:gridfit is faster by 5.67x       Construct:gridfit is faster by 3.08x
Grid size:256,  Samples:500     Interp:gridfit is faster by 1.72x       Construct:gridfit is faster by 3.93x
Grid size:384,  Samples:500     Interp:gridfit is faster by 2.99x       Construct:gridfit is faster by 2.42x
Grid size:512,  Samples:500     Interp:gridfit is faster by 2.82x       Construct:gridfit is faster by 2.21x
Grid size:1024, Samples:500     Interp:gridfit is faster by 2.83x       Construct:gridfit is faster by 6.58x

Grid size:50,   Samples:1000    Interp:gridfit is faster by 2.34x       Construct:gridfit is faster by 1.53x
Grid size:100,  Samples:1000    Interp:gridfit is faster by 2.23x       Construct:gridfit is faster by 4.11x
Grid size:200,  Samples:1000    Interp:gridfit is faster by 2.06x       Construct:gridfit is faster by 5.48x
Grid size:256,  Samples:1000    Interp:gridfit is faster by 1.97x       Construct:gridfit is faster by 5.76x
Grid size:384,  Samples:1000    Interp:gridfit is faster by 1.79x       Construct:gridfit is faster by 8.98x
Grid size:512,  Samples:1000    Interp:gridfit is faster by 1.75x       Construct:gridfit is faster by 7.85x
Grid size:1024, Samples:1000    Interp:scipy is faster by 1.10x Construct:gridfit is faster by 3.80x

Grid size:50,   Samples:10000   Interp:gridfit is faster by 1.49x       Construct:gridfit is faster by 3.18x
Grid size:100,  Samples:10000   Interp:gridfit is faster by 1.46x       Construct:gridfit is faster by 3.27x
Grid size:200,  Samples:10000   Interp:gridfit is faster by 1.54x       Construct:gridfit is faster by 3.36x
Grid size:256,  Samples:10000   Interp:gridfit is faster by 1.27x       Construct:gridfit is faster by 3.29x
Grid size:384,  Samples:10000   Interp:gridfit is faster by 1.45x       Construct:gridfit is faster by 2.87x
Grid size:512,  Samples:10000   Interp:gridfit is faster by 1.32x       Construct:gridfit is faster by 4.78x
Grid size:1024, Samples:10000   Interp:gridfit is faster by 1.65x       Construct:gridfit is faster by 18.46x

Grid size:50,   Samples:100000  Interp:gridfit is faster by 1.67x       Construct:gridfit is faster by 3.55x
Grid size:100,  Samples:100000  Interp:gridfit is faster by 1.49x       Construct:gridfit is faster by 3.46x
Grid size:200,  Samples:100000  Interp:gridfit is faster by 1.26x       Construct:gridfit is faster by 2.32x
Grid size:256,  Samples:100000  Interp:gridfit is faster by 1.31x       Construct:gridfit is faster by 2.78x
Grid size:384,  Samples:100000  Interp:gridfit is faster by 1.17x       Construct:gridfit is faster by 3.25x
Grid size:512,  Samples:100000  Interp:gridfit is faster by 1.61x       Construct:gridfit is faster by 4.60x
Grid size:1024, Samples:100000  Interp:gridfit is faster by 1.47x       Construct:gridfit is faster by 4.23x