Introduction To NumPy

Installing NumPy and other packages using conda

In addition to numpy some of the examples use matplotlib and scipy for plotting and fitting.

To install numpy and other python related packages I recommend the conda package manager which will install python too. This does not interfere with any preexisting python on your system.

• https://conda.io/docs/
• https://conda.io/docs/user-guide/install/index.html

After installing miniconda3, you can:

conda install numpy ipython matplotlib     ## bare minimum
conda install scipy                        ## fitting + much more
conda install sympy                        ## symbolic maths

All the most popular machine learning packages are also available via conda.

NumPy Documentation

• http://www.numpy.org

See Also

• https://simoncblyth.github.io/env/presentation/opticks_gpu_optical_photon_simulation_sep2018_qingdao.html

  Presention slides in html, with many NumPy examples. Once the images load (it takes a while sometimes) the html provides a javascript interface to navigate the slide pages, with menu at bottom right.

• https://simoncblyth.github.io/env/presentation/opticks_gpu_optical_photon_simulation_sep2018_qingdao.txt

  If your connection to github is slow, you can get the RST text sources of the slides at this URL

• https://github.com/simoncblyth/intro_to_cuda/

  Learning NumPy is ideal preparation for Learning CUDA...

Hello World Examples

This repository contains a few very simple examples of using NumPy.

• https://github.com/simoncblyth/intro_to_numpy/

Start by cloning the repository to a machine with NumPy installed:

which git    # Git is required
git clone https://github.com/simoncblyth/intro_to_numpy

If you are planning to contribute and get write permission cloning from the ssh url is more convenient:

git clone git@github.com:simoncblyth/intro_to_numpy.git

load/save NumPy arrays from C++

• https://github.com/simoncblyth/np

Highlighted Examples

recon.py

• generate arrival times at coordinates on a sphere of a "disturbance" starting within the sphere, assuming a normal distribution around geometric time
• perform NLL minimization to find "disturbance" origin position parameters from the arrival times at positions on the sphere
• positions, times and params are saved as npy files, for use by the recon extended example in https://github.com/simoncblyth/intro_to_cuda/tree/master/recon

recon_terse.py

terse version of recon.py demonstrating array-oriented brevity

accept_reject_sampling.py

demo technique with plot

estimate_pi.py

MC method estimate

exponential_pdf_cdf.py

use of matplotlib to make a simple plot

header.py

use np.fromfile to examine the NPY header of an array and extract the metadata dict

random_persist.py

demonstrate random generation and persisting

structured.py

investigate numpy structured arrays, with composite dtype

normal.py

checking scipy.stats.norm

dydx.py

incomplete try at comparing numpy to python performance

python_vs_numpy_vs_cupy/ellipse_closest_approach_to_point.py

compare python to numpy and cupy, in this example:

• numpy is factor 10 faster than pure python
• cupy is more than a factor 1000 faster than numpy (NVIDIA TITAN V)

NumPy Introductions

Python Ecosystem

• http://www.scipy-lectures.org/index.html
• http://www.scipy-lectures.org/intro/intro.html
• http://www.scipy-lectures.org/intro/numpy/index.html

Informative Papers on NumPy

The NumPy array: a structure for efficient numerical computation Stefan Van Der Walt, S. Chris Colbert, Gaël Varoquaux

• https://hal.inria.fr/inria-00564007/document

Slides

• https://github.com/ContinuumIO/tutorials/blob/master/Intro_to_NumPy.pdf