numba-vectorization

This repository aims to provide different analyses about the functionalities that numba vectorize brings. This typically bring extra plots, metrics and parametrization.

Setup

Before running anything, install the package with:

pip install .

This will guarantee that all dependencies are solved.

Current implemented tools

Universal functions with @vectorize

• numba_vectorization\vectorize_analysis.py: Perform an analysis that evaluates the performance of different techniques with vectorized functions. Arguments:
  • --print-metrics: True to print all resulting metrics.
  • --plot-metrics: True to plot all metrics based on number of samples used.
  • --n-calls: Choose how many times each function is evaluated to perform its average.
  • --up-to-n-samples: Indicates maximum number of samples to be processed.
  • --increase-by: Rate of increase in the sample size used in each iteration.
• numba_vectorization\vectorize_ufunc_features.py: Check how a vectorized function with numba acquire extra features thanks to numpy ufuncs.

Generalized universal functions with @guvectorize

• numba_vectorization\simple_guvectorize_analysis.py: Perform an analysis that evaluates the performance of different techniques with simple guvectorized function. Arguments:
  • --print-metrics: True to print all resulting metrics.
  • --plot-metrics: True to plot all metrics based on number of samples used.
  • --n-calls: Choose how many times each function is evaluated to perform its average.
  • --up-to-n-samples: Indicates maximum number of samples to be processed.
  • --increase-by: Rate of increase in the sample size used in each iteration.
• numba_vectorization\fk_guvectorize_analysis.py: Perform an analysis that evaluates the performance of different techniques with complex guvectorized functions. These functions perform forward kinematics equation for an arbitrary robot with 6 joints. Arguments:
  • --print-metrics: True to print all resulting metrics.
  • --plot-metrics: True to plot all metrics based on number of samples used.
  • --n-calls: Choose how many times each function is evaluated to perform its average.
  • --up-to-n-samples: Indicates maximum number of samples to be processed.
  • --increase-by: Rate of increase in the sample size used in each iteration.