====== Numpy ======

===== openmp, mkl 쓰레드 개수 조정 =====

openmp, mkl에서 1 쓰레드만 사용하도록 하려면,
환경변수 OMP_NUM_THREADS, MKL_NUM_THREADS를 수정 해야함

cmd 창에서
<code>
set OMP_NUM_THREADS=1
set MKL_NUM_THREADS=1
</code>

또는, python에서
<code python>
os.environ['OMP_NUM_THREADS'] = '1'
os.environ['MKL_NUM_THREADS'] = '1'
</code>

<code>
export CUDA_VISIBLE_DEVICES=""
export MKL_DEBUG_CPU_TYPE=5
export MKL_SERIAL=YES; export OMP_NUM_THREADS=1
</code>

  * http://publicclu2.blogspot.com/2013/05/intel-complier-suite-reference-card.html


===== log =====

  * 임의의 log base 사용

<code python>
import math
import numpy as np

# base: 42 사용
math.log(10, 42) == np.log(10) / np.log(42)
</code>


===== NaN 처리 =====

  * https://pypi.org/project/Bottleneck/

===== 관련 도구 =====

  * https://cupy.chainer.org/
  * https://numba.pydata.org/