Using RStudio and the reticulate package to benchmark important algorithms across R and Python environments
Environment: Ubuntu 18 openblas R 3.6.3 python 3.7.4 i7 6700HQ 24G DDR4 2133
sudo apt-get install libopenblas-dev
Hypothesis: There is no significant gap in an algorithms performance between R and Python environments.
Algebra operations, OLS and in development phase RandomForest,random numbers generators (Mersenne-Twister, ...), NN and Xgb
Proc usage was calculated by R function microbenchmark::get_nanotime and Python method time.clock
X'X takes around mn^2 flops ( the 2 power were reduced during last years but it will be always exponential).
So time growing linear with additional rows and quadratic with additional columns.
XB=Y - mostly it is a overdetermined system
or X'XB=X'Y - where X'X is a square matrix
OLS consists of solving linear systems with usage of cholesky,QR or LU matrix decomposition
Example under QR decomposition:
