About

This is material I prepared for a class. I was too ambitious.

I didn't have time to show numpy/scipy. I didn't show json-files. I didn't show virtualenvs. You should look into these things, they are pure awesomeness.

Files

• 1.py: Shows scoping by indentation. Python's most discussion-about-this-is-futile-but-fun feature.
• 2.py: Basic lists, loops, tests, and scope again.
• 3.py: Duck-typing: OO sans interfaces and inheritance! IMO C++ templates done right :) (Of course I understand it's a dynamic language. No. Stop it.)
• ipython_log.py: The log of my live-session, analyzing some CSV data. I added comments to it.
• test.py: This will lead to an obvious-in-hindsight, eye-opening realization at some point in your coder's life. Maybe in the future, maybe in the past, maybe now.

Preparations

Create (and activate) a "virtual environment", this installs all python-stuff into one directory so you don't mess with your system.

$ virtual_env env
$ . env/bin/activate

Install the ipython console and its necessary dependencies. IPython is more user-friendly than the regular python shell

$ pip install ipython
$ easy_install readline

Numpy & scipy

Numpy brings performant matrices (called ndarrays) to python. It also includes a GEMM&friends-wrapper called "numpy.dot". Scipy can be compared to a wrapper around LAPACK, FFTPACK and many more.

$ pip install numpy
$ pip install scipy

Note that it is possible, but tedious, to link numpy against a HP-BLAS implementation like OpenBLAS, MKL, ACML and friends. I am still working on a full tutorial for installing numpy+scipy+OpenBLAS. In the meantime, instructions for a global install using MKL are provided by Intel.

Check to which BLAS numpy was linked

If you need to veriy which BLAS implementation your numpy is using, write the following in a python console:

import numpy.distutils.system_info as sysinfo
sysinfo.get_info('blas')

This gives you the name of the linked library. You could then ldd /usr/lib/libblas.so (if that is the indicated file) to get a few more details.

Plotting

matplotlib is a plotting library which is very similar to matlab's plotting. I will not introduce it, because I don't like/use matlab's plotting. I like d3.js.

General python introduction

• %logstart is like matlab's diary command. Commands starting with a %-sign are IPython "magic" commands.

General stuff

These are my "lecture notes", i.e. what I didn't want to forget to tell you guys.

Two camps of sci

• Compute compute compute, using known algos

  • Paolo, physicists, chemists

• Discover, discover, discover

  • "Big data", Biologists, Data/Social/Financial Analysts
  • Data transformation everywhere
  • Use many tools

• Of course not B/W

What's Python?

• Scripting language from the 90s (read: young)
• GvR is the BDFL (Benevolent Dictator For Life)
• Readable, practical, ... "import this"
• "Dump brain to editor" - Some guy on the internet

When not to use Python

• High numerical performance is central
• Big GUI code
• Windows only
• You are an IDE hugger
• You are a static typing lover

When to use Python

• All other cases :)
• Have fun programming
• Glue code
• Prototypes
• Fast development time

Uncleaned notes

Any other notes I had made quickly but didn't have the time to "cleanup".

TODO:

• ipython: %logon/logstart _ _i
• Indentation-scoped
• Readable (professors = ['Bientinesi', 'Leibe', 'Lichter'] if 'lucas' not in professors: print('No permission') )
• dynamic typing
  • Still has types: int (c:long), long (c:mpl), float (c:double), complex
• OO+Duck-typing
• numpy?
  • a[x>5]
• Away from OO->dicts+lists(+tuples(+sets))
• (immutability: strings, tuples)
• To functional: map, filter, all/any
• To pythonic: list/dict comprehension
• [bla for i for j] # j is faster
• generators/lazyness qsort1 = lambda lst : lst if len(lst) <= 1 else qsort1([i for i in lst[1:] if i < lst[0]]) + [lst[0]] + qsort1([i for i in lst[1:] if i >= lst[0]])