ml_perf_test
Utilities and notebooks with some performance tests for typical ML tasks
wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
sh Miniconda3-latest-Linux-x86_64.sh
. ~/miniconda3/etc/profile.d/conda.sh
conda create --name testmkl pip numpy scipy scikit-learn joblib
conda activate testmkl
conda install -c conda-forge lightgbm
To make sure you use packages based on Intel MKL, run python shell, import numpy and then check what libs are loaded, the list should contain some mkl libs
For example:
lsof -p 5971 | grep mkl
where 5971 is PID of your python shell process.
As an option, you can also create a no mkl enviroment in the following way:
conda create --name testnomkl python=3.6 pip
pip install numpy scipy scikit-learn lightgbm
Run
python ml_text_bench.py