Getting started

• Download and install python
• Options for running code:
  1. Write script first, and then execute it from a terminal; or,
  2. Execute scripts/commands from interactive console or GUI.

Running a script from terminal

Running a script from terminal

Interactive console: IPython

Integrated Development Environment (IDE): Spyder

NumPy

• Multidimensional array objects of various data types (floating point, integer, bool, etc.)
• Array slicing / broadcasting
• Linear algebra (basic ones), Fourier transform, random number generation tools
• Tools for integrating Fortran code
• NumPy guide for Matlab users

NumPy example

SciPy

• scipy.cluster - Hierarchical clustering
• scipy.integrate - Integration (uni- and multi-dimensional) and ODE solvers
• scipy.interpolate - Interpolation (uni- and multi-dimensional, splines, etc.)
• scipy.io - Various file readers / writers (including Matlab format)
• scipy.linalg - Linear algebra routines
• scipy.ndimage - Image processing routines (filters, interpolation, i/o, morphology, etc.)
• scipy.optimize - Optimization (too many methods to list!)
• scipy.signal - DSP routines (filters, wavelets, spectrogram analysis)
• scipy.sparse - Sparse matrix data structures and graph routines
• scipy.special - Various special functions (Bessel functions, orthogonal polynomials, etc.)
• scipy.stats - Statistical tests, pdf parameter fitting, Gaussian KDE, etc.

• 2D / 3D plotting
• Renders $\rm\LaTeX$
• Animation writers - gif / mpeg
• Interactive GUI stuff - sliders, buttons, etc.

Compiling Python code to C - Cython

• Involves typing variables and function return types / arguments
• Sometimes useful to speed up code after it has been prototyped in Python
• Must be careful to type *all* variables, or code may still be slow

Cython example

Cython example

Cython example

# Creates ip_cython.c file.
cython ip_cython.pyx 

# Creates ".so" file that can be imported as Python module.
gcc -shared -pthread -fPIC -fwrapv -O2 -Wall \
-fno-strict-aliasing -I/usr/include/python2.7 \
-o ip_cython.so ip_cython.c 
                        

Cython example

Calling Fortran from Python - f2py

• Utility that comes packaged with NumPy
• Creates Python wrappers for Fortran modules/subroutines/functions
• Numerical variables / NumPy arrays are passed through Python wrapper function to Fortran routines where treated as native Fortran objects

f2py example

f2py -c -m ip_fortran ip_fortran.f90
                        

f2py example

Machine learning - Scikit-learn

aka "sklearn"

• Classification - SVM, Logistic regression, Random Forests, neural networks, etc.
• Regression - SVR, Random Forest Regression, ridge, LASSO, etc.
• Clustering - K-means, spectral clustering, etc.
• Dimensionality reduction - PCA, kernel PCA, Laplacian Eigenmaps, etc.
• Model selection - Grid search, cross validation, performance metics
• Pre-processing - Filling missing values, encoding labels, etc.

Sklearn example - classification with SVM

Sklearn example - classification with SVM

python ./sklearn-example.py
                        

Symbolic Math - SymPy

• Used by Sage Math
• Calculus
• Combinatorics
• Equation solving
• Polynomials

Sympy example

Symbolic computation chains, symbolic differentiation, CPU/GPU compilation - Theano

• Create computation chains symbolically
• Symbolic differentiation
• Compiles symbolic Python computation chains to CPU/GPU code
• Used a lot by neural network / "deep learning" folks, but not a neural-network specific library, e.g. Caffe

Matrix multiplication - Theano vs. Numpy / GPU vs. CPU

Theano example - grad. descent with symbolic gradient computation

Theano example - grad. descent with symbolic gradient computation

Theano example - grad. descent with symbolic gradient computation

• More advanced image-processing methods (than, e.g., scipy.ndimage
• Low-level ops: exposure, histogram equalization, edge detection
• Mesh generation: marching cubes
• Object detection: Hough transforms, corner/blob detection, template matching
• Gabor filtering / texture analysis
• Lots of filtering: deconvolution, denoising, in-painting
• Segmentation: active contours, random walker, watershed, otsu thresholding

• Generated using skimage's marching cubes method on boolean volume.
• Meshed surface plotted using Matplotlib