No performance gain observed on kmeans scikit-learn with tbb and mkl based intelnumpy

[vineel96]

Hi @anton-malakhov @rscohn2 @samaid @sfblackl-intel , I did not observe any performance gain in execution time on scikit-learn's kmeans algorithm with tbb monkey patching and intelnumpy(which uses MKL) over normal kmeans algorithm. system: aws c6i.4xlarge icelake with 8 cores(2 threads per core) I have installed intelnumpy, tbb, smp, numba. I have following doubts:

1. do TBB monkey patching mimics 100% of oneTBB which is used in onedal(scikit-learn-intelex)?
2. Based on results can we conclude that oneMKL and oneTBB does not contribute at all for performance gain for kmeans which are major building blocks in scikit-learn-intelex?