why-not/numpy_testing

Numpy Testing

the purpose of this repository is to learn more about LAPACK, BLAS and ATLAS.

How to find the location of LAPACK, BLAS, ATLAS

one can use locate command (https://openmx.ssri.psu.edu/wiki/blas-library-help)

locate libblas.so

another way to get information about lapack, blas, atlas. liblapack3, libblas3 , libatlas-base-dev(https://stackoverflow.com/questions/15851831/how-to-check-if-blas-and-atlas-already-installed)

apt-cache policy liblapack3
apt-cache policy libblas3
apt-cache policy libatlas-base-dev

currently I don't have them installed

(base) femtoland@femtoland:~$ apt-cache policy liblapack3
liblapack3:
  Installed: (none)
  Candidate: 3.7.1-4ubuntu1
  Version table:
     3.7.1-4ubuntu1 500
        500 http://us.archive.ubuntu.com/ubuntu bionic/main amd64 Packages
(base) femtoland@femtoland:~$ apt-cache policy libblas3
libblas3:
  Installed: (none)
  Candidate: 3.7.1-4ubuntu1
  Version table:
     3.7.1-4ubuntu1 500
        500 http://us.archive.ubuntu.com/ubuntu bionic/main amd64 Packages
(base) femtoland@femtoland:~$ apt-cache policy libatlas-base-dev
libatlas-base-dev:
  Installed: (none)
  Candidate: 3.10.3-5
  Version table:
     3.10.3-5 500
        500 http://us.archive.ubuntu.com/ubuntu bionic/universe amd64 Packages

Test on Mac Book Pro

Hardware information

Model Name:	MacBook Pro
Model Identifier:	MacBookPro14,1
Processor Name:	Intel Core i5
Processor Speed:	2.3 GHz
Number of Processors:	1
Total Number of Cores:	2
L2 Cache (per Core):	256 KB
L3 Cache:	4 MB
Hyper-Threading Technology:	Enabled
Memory:	16 GB
Boot ROM Version:	198.0.0.0.0
SMC Version (system):	2.43f6

Results of tests

n [33]: run test.py
read 32 images 1 times: 2.53048706055 seconds per loop
sum axis=0 32 images 10 times: 0.612431001663 seconds per loop
sum axis=1 32 images 10 times: 0.39970459938 seconds per loop
sum full 32 images 10 times: 0.343644690514 seconds per loop
mean 32 images 10 times: 0.324725699425 seconds per loop

In [34]: run test.py
read 32 images N = 5 times: 3.28182039261 seconds per loop
sum axis=0 32 images N = 20 times: 0.698462152481 seconds per loop
sum axis=1 32 images N = 20 times: 0.39868710041 seconds per loop
sum full 32 images N = 20 times: 0.343178701401 seconds per loop
mean 32 images N = 20 times: 0.333559799194 seconds per loop

In [1]: run test.py
read 32 images N = 5 times: 2.56678800583 seconds per loop
sum axis=0 32 images N = 20 times: 0.611644756794 seconds per loop
sum axis=1 32 images N = 20 times: 0.397477507591 seconds per loop
sum full 32 images N = 20 times: 0.341990947723 seconds per loop
mean 32 images N = 20 times: 0.320617651939 seconds per loop

Test on AWS Ubuntu Image optimized for Deep Learning. (t3a.2xlarge: 8 cores, 32 GB Ram)

run test.py
generate_data, N times = 50 times: 0.05521430158000044 seconds per loop
sum axis=0 N times= 200 times: 0.013160574044999863 seconds per loop
sum axis=1 N times = 200 times: 0.013259274245000086 seconds per loop
sum full N times= 200 times: 0.01330458675000017 seconds per loop
mean 32 N times = 200 times: 0.01342332953500005 seconds per loop

run test.py 
generate_data, N times = 50 times: 0.05374531482000066 seconds per loop
sum axis=0 N times= 200 times: 0.011327756174999877 seconds per loop
sum axis=1 N times = 200 times: 0.013209703784999647 seconds per loop
sum full N times= 200 times: 0.013074761825000109 seconds per loop
mean 32 N times = 200 times: 0.013217427714999985 seconds per loop

run test.py 
generate_data, N times = 50 times: 0.05423383809999905 seconds per loop
sum axis=0 N times= 200 times: 0.011368431889999897 seconds per loop
sum axis=1 N times = 200 times: 0.013211888019999946 seconds per loop
sum full N times= 200 times: 0.013100763045000009 seconds per loop
mean 32 N times = 200 times: 0.013218856825000102 seconds per loop

Test on AWS Ubuntu Image optimized for Deep Learning. (c5d.2xlarge: 8 cores, 16 GB Ram)

ubuntu@ip-172-30-0-9:~$ source activate mxnet_p36
(mxnet_p36) ubuntu@ip-172-30-0-9:~$ ipython
Python 3.6.5 |Anaconda, Inc.| (default, Apr 29 2018, 16:14:56)
Type 'copyright', 'credits' or 'license' for more information
IPython 6.4.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: run test.py
generate_data, N times = 50 times: 0.04426059433999995 seconds per loop
sum axis=0 N times= 200 times: 0.009342705459999934 seconds per loop
sum axis=1 N times = 200 times: 0.013293346549999967 seconds per loop
sum full N times= 200 times: 0.013351381415000106 seconds per loop
mean 32 N times = 200 times: 0.009968762909999923 seconds per loop

In [2]: run test.py
generate_data, N times = 50 times: 0.04374865192000016 seconds per loop
sum axis=0 N times= 200 times: 0.009376861555000034 seconds per loop
sum axis=1 N times = 200 times: 0.012888064189999967 seconds per loop
sum full N times= 200 times: 0.012832252030000007 seconds per loop
mean 32 N times = 200 times: 0.009922239149999967 seconds per loop

In [3]: run test.py
generate_data, N times = 50 times: 0.04357514243999958 seconds per loop
sum axis=0 N times= 200 times: 0.009447478594999979 seconds per loop
sum axis=1 N times = 200 times: 0.013025719820000035 seconds per loop
sum full N times= 200 times: 0.013008963770000008 seconds per loop
mean 32 N times = 200 times: 0.009978990769999996 seconds per loop

Test on AWS Ubuntu Image optimized for Deep Learning. (t2.nano: 1 cores, 0.5 GB Ram, Smallest Possible Amazon Server)

In [1]: run test.py
generate_data, N times = 50 times: 0.06553657586 seconds per loop
sum axis=0 N times= 200 times: 0.013732203115000025 seconds per loop
sum axis=1 N times = 200 times: 0.016420330545000007 seconds per loop
sum full N times= 200 times: 0.01639490290000001 seconds per loop
mean 32 N times = 200 times: 0.015590667734999961 seconds per loop

In [2]: run test.py
generate_data, N times = 50 times: 0.06461507319999982 seconds per loop
sum axis=0 N times= 200 times: 0.01365810035999992 seconds per loop
sum axis=1 N times = 200 times: 0.016625747829999968 seconds per loop
sum full N times= 200 times: 0.01636717886999989 seconds per loop
mean 32 N times = 200 times: 0.0157351610149999 seconds per loop

In [3]: run test.py
generate_data, N times = 50 times: 0.06342938600000025 seconds per loop
sum axis=0 N times= 200 times: 0.013575194484999997 seconds per loop
sum axis=1 N times = 200 times: 0.01668353084000003 seconds per loop
sum full N times= 200 times: 0.016531246105000008 seconds per loop
mean 32 N times = 200 times: 0.015784036969999988 seconds per loop

Information About Installed Linear Algebra Accelerators

(Same for all the amazon tests)

apt-cache policy liblapack3
liblapack3:
  Installed: 3.6.0-2ubuntu2
  Candidate: 3.6.0-2ubuntu2
  Version table:
 *** 3.6.0-2ubuntu2 500
        500 http://us-east-1.ec2.archive.ubuntu.com/ubuntu xenial/main amd64 Packages
        100 /var/lib/dpkg/status
(mxnet_p36) ubuntu@ip-172-30-0-9:~$ apt-cache policy libblas3
libblas3:
  Installed: 3.6.0-2ubuntu2
  Candidate: 3.6.0-2ubuntu2
  Version table:
 *** 3.6.0-2ubuntu2 500
        500 http://us-east-1.ec2.archive.ubuntu.com/ubuntu xenial/main amd64 Packages
        100 /var/lib/dpkg/status
(mxnet_p36) ubuntu@ip-172-30-0-9:~$ apt-cache policy libatlas-base-dev
libatlas-base-dev:
  Installed: 3.10.2-9
  Candidate: 3.10.2-9
  Version table:
 *** 3.10.2-9 500
        500 http://us-east-1.ec2.archive.ubuntu.com/ubuntu xenial/universe amd64 Packages
        100 /var/lib/dpkg/status

How to disengage LAPACK, BLAS, ATLAS on mac and linux

I have asked https://apple.stackexchange.com/questions/368380/lapack-blas-atlas-on-mac-book-pro-2017-mojave

references:

• [LAPACK] https://en.wikipedia.org/wiki/LAPACK
• [BLAS] http://www.netlib.org/blas/
• [ATLAS] https://en.wikipedia.org/wiki/Automatically_Tuned_Linear_Algebra_Software