iPython
The SciPy Stack
SciPy is a Python-based ecosystem of libraries and tools for scientific computing and data analytics
iPython is the primary way of interacting with the SciPy stack -- whether through the shell or a Jupyter notebook. It's also a fantastic REPL.
iPython
Two modes:
-
Interactive shell
- Replacement for
pythonREPL
- Replacement for
-
Jupyter notebook
- Interactive web-based documents mixing text, executable code, graphics
Before we proceed, make sure your computer is ready (OS shell):
pip install ipython
iPython Shell History
In [1]: ['Sage', 'Thyme', 'Oragano', 'Posh']
Out[1]: ['Sage', 'Thyme', 'Oragano', 'Posh']
In [2]: type(In[1])
Out[2]: str
In [3]: type(Out[1])
Out[3]: list
In [4]: spices = Out[1]
In [5]: spices
Out[5]: ['Sage', 'Thyme', 'Oragano', 'Posh']
In [6]: spices is Out[1]
Out[6]: True
In is a list, Out is a dict.
iPython Help
Single ? gives abbeviated version of python's help
In [7]: def add(a, b):
...: """Return the result of + operation on a and b"""
...: return a + b
...:
In [8]: add?
Signature: add(a, b)
Docstring: Return the result of + operation on a and b
File: `/cs2316/<ipython-input-7-af5293282e78>
Type: function
Double ?? gives source code, if available.
In [9]: add??
Signature: add(a, b)
Source:
def add(a, b):
"""Return the result of + operation on a and b"""
return a + b
File: `/cs2316/<ipython-input-7-af5293282e78>
Type: function
Doctests
All functions should have doctests. But doctests must be in in Python REPL format. iPython has you covered.
In [10]: %doctest_mode
Exception reporting mode: Plain
Doctest mode is: ON
>>> add(1, 2)
3
>>> add("foo", "bar")
'foobar'
>>> %doctest_mode
Exception reporting mode: Context
Doctest mode is: OFF
In [14]:
Now you can add that to your function definition.
In [16]: def add(a, b):
...: """Return the result of + operation on a and b
...:
...: >>> add(1, 2)
...: 3
...: >>> add("foo", "bar")
...: 'foobar'
...: """
...: return a + b
...:
iPython Magic Commands
Special commands provided by iPython, prepended by %.
- Run a Python script from within iPython:
In [35]: %run people.py
[<Stan, 2008-08-13, 150cm, 45kg>,
<Kyle, 2008-02-25, 160cm, 50kg>,
<Cartman, 2008-05-26, 140cm, 100kg>,
<Kenny, 2009-07-30, 130cm, 40kg>]
- Get help with a magic command with
?
In [2]: %cd?
Docstring:
Change the current working directory.
(content elided)
Usage:
cd 'dir': changes to directory 'dir'.
(additional output elided)
Get a list of all magic commands with %lsmagic
iPython Shell Commands
Run shell commands by prepending with a !
In [27]: !ls *.py
fun.py grades.py maths.py people.py pp.py
In [28]: pyscripts = !ls *.py
In [29]: pyscripts
Out[29]: ['fun.py', 'grades.py', 'maths.py', 'people.py', 'pp.py']
iPython provides magic commands for most common shell commands.
iPython Directory Bookmarking
Great timesaving feature: bookmark directories
In [3]: %pwd
Out[3]: '/home/chris/vcs/github.com/cs2316/cs2316.github.io/code'
In [4]: %cd
/home/chris
In [5]: %bookmark cs2316code `chris/vcs/github.com/cs2316/cs2316.github.io/code
In [6]: cd cs2316code
(bookmark:cs2316code) -> `chris/vcs/github.com/cs2316/cs2316.github.io/code
/home/chris/vcs/github.com/cs2316/cs2316.github.io/code
iPython Automagic commands
With automagic turned on, some shell commands can be run as if they were built into iPython:
In [22]: pwd
Out[22]: '/Users/chris/cs2316'
In [23]: ls *.py
fun.py grades.py maths.py people.py pp.py
-
Toggle automagic on and off with
%automagic. -
These commands work with automagic (list them yourself iwth
%lsmagic):%cd,%cat,%cp,%env,%ls,%man,%mkdir,%more,%mv,%pwd,%rm, and%rmdir
Timing Code in iPython
In [23]: import numpy as np
In [24]: pylist = list(range(1, 100000))
In [25]: nparray = np.arange(1, 1000000)
In [35]: %timeit _ = [x * 2 for x in pylist]
1.41 ms ± 15.1 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
In [37]: %timeit _ = nparray * 2
620 μs ± 17.5 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
Numpy is a little faster than native Python. (To be fair, Python now has an efficient array type but by the time it was added to the Python standard library, everyone was already using and depending on Numpy.)
Profiling a Script
In [7]: %run -p -l 10 -s cumulative funcalc.py
2673375 function calls (1147466 primitive calls) in 1.691 seconds
Ordered by: cumulative time
List reduced from 56 to 10 due to restriction <10>
ncalls tottime percall cumtime percall filename:lineno(function)
2/1 0.000 0.000 1.691 1.691 {built-in method builtins.exec}
1 0.000 0.000 1.691 1.691 <string>:1(<module>)
1 0.000 0.000 1.691 1.691 interactiveshell.py:2431(safe_execfile)
1 0.000 0.000 1.691 1.691 py3compat.py:182(execfile)
1 0.000 0.000 1.690 1.690 funcalc.py:1(<module>)
1 0.000 0.000 1.689 1.689 funcalc.py:46(main)
1 0.039 0.039 1.689 1.689 funcalc.py:34(profile)
510961/10000 0.510 0.000 0.603 0.000 funcalc.py:14(sub)
510961/10000 0.514 0.000 0.598 0.000 funcalc.py:6(add)
510961/10000 0.340 0.000 0.340 0.000 funcalc.py:22(mult)
-pmeans profile-l 10means show only 10 lines-s cumulativemeans sort by cumulative time
Profiling a Function
%prun profiles a function. Uses same options as %run -p.
In [10]: %prun -l 10 -s cumulative funcalc.profile()
2673429 function calls (1148052 primitive calls) in 1.726 seconds
Ordered by: cumulative time
List reduced from 15 to 10 due to restriction <10>
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.000 0.000 1.726 1.726 {built-in method builtins.exec}
1 0.000 0.000 1.726 1.726 <string>:1(<module>)
1 0.042 0.042 1.726 1.726 funcalc.py:34(profile)
511231/10000 0.537 0.000 0.620 0.000 funcalc.py:6(add)
511231/10000 0.523 0.000 0.615 0.000 funcalc.py:14(sub)
511231/10000 0.336 0.000 0.336 0.000 funcalc.py:22(mult)
20000 0.019 0.000 0.097 0.000 random.py:223(randint)
501231 0.092 0.000 0.092 0.000 funcalc.py:15(dec)
501231 0.082 0.000 0.082 0.000 funcalc.py:7(inc)
20000 0.036 0.000 0.078 0.000 random.py:179(randrange)