# Python I/O: Reading, Writing, Pickling¶

Python has several built-in tools for handling files. Let's cover the most important ones.

## File Objects and ASCII¶

ASCII files are handled like everything else- with objects. Specifically, file objects. These objects contain all you need to open, close, and write files. The file used for this excercise is "player_rankings.txt". It is a list of players eligible for the 2011 draft sorted by their ranking (i.e., how draft pundits rated the players). Information such as position and school are also included. Make sure you have this file in your current working directory before running this notebook!

In [2]:
# Create a file object
f = open('player_rankings.txt', 'r')
# Attributes tell us things about the file.
print(f.mode)
print(f.name)
# Methods perform the usual file actions.
print(a)

r
player_rankings.txt
1.	Patrick Peterson	CB	L.S.U.



Per the usual, help or IPython's ? operator combined with tab-complete will help you explore the object's methods easily.

Above, we first opened the file and created a file object, f. The file name is given as a string (literal or variable.) We opened the file in read mode via the 'r' argument. It's also possible to open in append mode ('a'), write mode ('w'), and various combinations (e.g. 'r+' for combined read-write.)

Next, we used the method readline to read a single line (until the invisible newline character is found) and move the file pointer to the position just after the newline character. Note that newline characters are platform dependent, but modern languages typically do the work for you. This method returned the read line as a string.
What if we wanted to read a bunch of lines at once?

In [3]:
lines = f.readlines()
print(lines[0:3])
f.close()

['     2.\tA.J. Green\tWR\tGeorgia\n', '     3.\tMarcell Dareus\tDT\t Alabama\n', '     4.\tVon Miller\tOLB\tTexas A&M\n']


Here, we see that the readlines method slurps the rest of the file from the current pointer location to the end. Then, the contents are separated by newline character and put into a list.

Like good file stewards, we closed the file using close. Python is very forgiving about leaving files open, but let's not take advantage of that good nature.

Note that file objects are iterators with a next method. This means, similar to list objects, we can put them into a for loop with minimal work. Let's loop through our file and find all of the players that are quarterbacks.

In [4]:
f = open('player_rankings.txt' ,'r')
for line in f:
if "QB" in line:
print(line.strip())
f.close()

9.	Blaine Gabbert	QB	Missouri
26.	Jake Locker	QB	Washington
29.	Cam Newton	QB	Auburn
38.	Ryan Mallett	QB	Arkansas
49.	Christian Ponder	QB	Florida St.
80.	Ricky Stanzi	QB	Iowa
98.	Colin Kaepernick	QB	Nevada
100.	Andy Dalton	QB	T.C.U.


Cam Newton ranked 29th overall? Preposterous!

Writing to files is just as easy:

In [5]:
f1 = open('player_rankings.txt', 'r')
f2 = open('test.txt', 'w')
f2.write('Ranked players that are position {0}'.format('QB') )
for line in f1:
if "QB" in line:
f2.write(line)
f1.close()
f2.close()


Note that when we use print, we often get a newline automatically. However, the write method is very explicit: if we don't add newlines, they won't show up.

In [24]:
# Look at the file created by this code:
f=open('test2.txt', 'w')
f.write('This line and')
f.write(' this one will be combined...\n')
f.write('But this one stands alone!\n')
f.close()

In [27]:
%more test2.txt
# IPython's "magic" commands let you perform command-line functions.
### IMPORTANT NOTE: The "more" command sometimes fails in Jupyter Notebooks... try this in a regular terminal.


Per the usual, use interactive help and IPython's auto-complete to explore the other file object methods.

Binary files won't be covered here, but they're not that different. Simply add b, for binary, to the open mode (e.g. 'rb' to write to a binary file.) Typically, however, the binary format you want to open is some specialized format that has a python package that will do the work for you.

## Pickles¶

Often, it is nice to dump data to a file without worrying about the format, especially if you can write objects to file and restore them in the exact same state. Python has such a mechanism: pickles. Pickling is the act of dumping objects directly to a special binary format that can be in a later session. Let's start by making a complicated object.

In [28]:
import numpy as np  # import Numpy; more on importing soon.
a={1:np.zeros(20), 2:np.arange(0, np.pi, np.pi/16.0), 3:['hot','dogs']}
for key in a:
print(a[key])

[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[0.         0.19634954 0.39269908 0.58904862 0.78539816 0.9817477
1.17809725 1.37444679 1.57079633 1.76714587 1.96349541 2.15984495
2.35619449 2.55254403 2.74889357 2.94524311]
['hot', 'dogs']


That is some stuff. We want to save this for a later python session. We could use the pickle module, but we'll use cpickle, which is much, much faster but syntactically equivalent.
Let's create a file, dump a pickle into it (don't say that aloud in public), and close it.

In [33]:
import pickle
f_out=open('temp.pkl','wb')  # Open a file in binary write mode.
pickle.dump(a, f_out)
f_out.close()


That was easy. Our data has been pickled! Let's load our pickle (again, don't say these things with people around.) Some critical notes here:

• You must open the output file in binary mode, hence the "wb" in the open command.
• In Python2.*, there are two "pickle" modules: "Pickle" and "cPickle". Always use "cPickle" in Python2.*
• We used the file extension ".pkl" here. It actually doesn't matter what you use, but I recommend using something that is recognizable in the future (i.e., ".pkl").
In [37]:
f_in = open('temp.pkl', 'rb') # Open a file in binary read mode.
b = pickle.load(f_in)       # Load the pickle into a variable.
print(b)
f_in.close()

{1: array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0.]), 2: array([0.        , 0.19634954, 0.39269908, 0.58904862, 0.78539816,
0.9817477 , 1.17809725, 1.37444679, 1.57079633, 1.76714587,
1.96349541, 2.15984495, 2.35619449, 2.55254403, 2.74889357,
2.94524311]), 3: ['hot', 'dogs']}


Huzzah! Note that we can pickle many objects into a single file. We then just call load once for each saved object, each time assigning it to a new variable.

In [40]:
c=['a','b','c','d']
d=np.matrix('1 2; 3 4')
f_out=open('temp2.pkl', 'wb')
pickle.dump(c, f_out)
pickle.dump(d, f_out)
f_out.close()
##### New Session ####
f_in=open('temp2.pkl', 'rb')
f_in.close()
print(e, g)

['a', 'b', 'c', 'd'] [[1 2]
[3 4]]


Note that if we try to load another pickled object from our file, an IOError exception is thrown, causing our program to crash. It's good to know how many pickles are in each file. Alternatively, it is possible to just save the items as a tuple or list.

In [42]:
f_out=open('temp3.pkl', 'wb')
pickle.dump( (a,b,c,d), f_out)
f_out.close()
##### New Session ####
f_in=open('temp3.pkl', 'rb')
w,x,y,z = pickle.load(f_in) # multiple assignment FTW!
f_in.close()
print(w,x,y,z)

{1: array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0.]), 2: array([0.        , 0.19634954, 0.39269908, 0.58904862, 0.78539816,
0.9817477 , 1.17809725, 1.37444679, 1.57079633, 1.76714587,
1.96349541, 2.15984495, 2.35619449, 2.55254403, 2.74889357,
2.94524311]), 3: ['hot', 'dogs']} {1: array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0.]), 2: array([0.        , 0.19634954, 0.39269908, 0.58904862, 0.78539816,
0.9817477 , 1.17809725, 1.37444679, 1.57079633, 1.76714587,
1.96349541, 2.15984495, 2.35619449, 2.55254403, 2.74889357,
2.94524311]), 3: ['hot', 'dogs']} ['a', 'b', 'c', 'd'] [[1 2]
[3 4]]


If you are given a pickle file and don't know how many objects have been pickle-dumped into it, this snippet of code helps:

In [46]:
f_in=open('temp2.pkl', 'rb')  # Open the file.
data = []  # Create an empty list.
while True:
try:

[['a', 'b', 'c', 'd'], matrix([[1, 2],