6.1.1. Working with Data Files¶

For our purposes, we will assume that our data files are text files–that is, files filled with characters. The Python programs that you write are stored as text files. We can create these files in any of a number of ways. For example, we could use a text editor to type in and save the data. We could also download the data from a website and then save it in a file. Regardless of how the file is created, Python will allow us to manipulate the contents.

In Python, we must open files before we can use them and close them when we are done with them. As you might expect, once a file is opened it becomes a Python object just like all other data. Table 1 shows the functions and methods that can be used to open and close files.

6.1.2. Finding a File on your Disk¶

Opening a file requires that you, as a programmer, and Python agree about the location of the file on your disk. The way that files are located on disk is by their path You can think of the filename as the short name for a file, and the path as the full name. For example on a Mac if you save the file hello.txt in your home directory the path to that file is /Users/yourname/hello.txt On a Windows machine the path looks a bit different but the same principles are in use. For example on windows the path might be C:\Users\yourname\My Documents\hello.txt

You can access files in sub-folders, also called directories, under your home directory by adding a slash and the name of the folder. For example, if you had a file called hello.py in a folder called CS150 that is inside a folder called PyCharmProjects under your home directory, then the full name for the file hello.py is /Users/yourname/PyCharmProjects/CS150/hello.py. This is called an absolute file path. An absolute file path typically only works on a specific computer. Think about it for a second. What other computer in the world is going to have an absolute file path that starts with /Users/yourname?

If a file is not in the same folder as your python program, you need to tell the computer how to reach it. A relative file path starts from the folder that contains your python program and follows a computer’s file hierarchy. A file hierarchy contains folders which contains files and other sub-folders. Specifying a sub-folder is easy – you simply specify the sub-folder’s name. To specify a parent folder you use the special .. notation because every file and folder has one unique parent. You can use the .. notation multiple times in a file path to move multiple levels up a file hierarchy. Here is an example file hierarchy that contains multiple folders, files, and sub-folders. Folders in the diagram are displayed in bold type.

Using the example file hierarchy above, the program, myPythonProgram.py could access each of the data files using the following relative file paths:

• data1.txt
• ../myData/data2.txt
• ../myData/data3.txt
• ../../otherFiles/extraData/data4.txt

Here’s the important rule to remember: If your file and your Python program are in the same directory you can simply use the filename like this: open('myfile.txt', 'r'). If your file and your Python program are in different directories then you must use a relative file path to the file like this: open('../myData/data3.txt', 'r').

6.1.3. Glossary¶

absolute file path

The name of a file that includes a path to the file from the root directory of a file system. An absolute file path always starts with a /.

relative file path

The name of a file that includes a path to the file from the current working directory of a program. An relative file path never starts with a /.

6.1.4. Reading a File the Imperative Way¶

As an example, suppose we have a text file called qbdata.txt that contains the following data representing statistics about NFL quarterbacks. Although it would be possible to consider entering this data by hand each time it is used, you can imagine that it would be time-consuming and error-prone to do this. In addition, it is likely that there could be data from more quarterbacks and other years. The format of the data file is as follows

First Name, Last Name, Position, Team, Completions, Attempts, Yards, TDs, Ints, Comp%, Rating

Colt McCoy QB CLE  135 222 1576    6   9   60.8%   74.5
Josh Freeman QB TB 291 474 3451    25  6   61.4%   95.9
Michael Vick QB PHI    233 372 3018    21  6   62.6%   100.2
Matt Schaub QB HOU 365 574 4370    24  12  63.6%   92.0
Philip Rivers QB SD    357 541 4710    30  13  66.0%   101.8
Matt Hasselbeck QB SEA 266 444 3001    12  17  59.9%   73.2
Jimmy Clausen QB CAR   157 299 1558    3   9   52.5%   58.4
Joe Flacco QB BAL  306 489 3622    25  10  62.6%   93.6
Kyle Orton QB DEN  293 498 3653    20  9   58.8%   87.5
Jason Campbell QB OAK  194 329 2387    13  8   59.0%   84.5
Peyton Manning QB IND  450 679 4700    33  17  66.3%   91.9
Drew Brees QB NO   448 658 4620    33  22  68.1%   90.9
Matt Ryan QB ATL   357 571 3705    28  9   62.5%   91.0
Matt Cassel QB KC  262 450 3116    27  7   58.2%   93.0
Mark Sanchez QB NYJ    278 507 3291    17  13  54.8%   75.3
Brett Favre QB MIN 217 358 2509    11  19  60.6%   69.9
David Garrard QB JAC   236 366 2734    23  15  64.5%   90.8
Eli Manning QB NYG 339 539 4002    31  25  62.9%   85.3
Carson Palmer QB CIN   362 586 3970    26  20  61.8%   82.4
Alex Smith QB SF   204 342 2370    14  10  59.6%   82.1
Chad Henne QB MIA  301 490 3301    15  19  61.4%   75.4
Tony Romo QB DAL   148 213 1605    11  7   69.5%   94.9
Jay Cutler QB CHI  261 432 3274    23  16  60.4%   86.3
Jon Kitna QB DAL   209 318 2365    16  12  65.7%   88.9
Tom Brady QB NE    324 492 3900    36  4   65.9%   111.0
Ben Roethlisberger QB PIT  240 389 3200    17  5   61.7%   97.0
Kerry Collins QB TEN   160 278 1823    14  8   57.6%   82.2
Derek Anderson QB ARI  169 327 2065    7   10  51.7%   65.9
Ryan Fitzpatrick QB BUF    255 441 3000    23  15  57.8%   81.8
Donovan McNabb QB WAS  275 472 3377    14  15  58.3%   77.1
Kevin Kolb QB PHI  115 189 1197    7   7   60.8%   76.1
Aaron Rodgers QB GB    312 475 3922    28  11  65.7%   101.2
Sam Bradford QB STL    354 590 3512    18  15  60.0%   76.5
Shaun Hill QB DET  257 416 2686    16  12  61.8%   81.3

To open this file, we would call the open function. The variable, fileref, now holds a reference to the file object returned by open. When we are finished with the file, we can close it by using the close method. After the file is closed any further attempts to use fileref will result in an error.

>>>fileref = open("qbdata.txt", "r")
>>>
>>>fileref.close()
>>>

The process of opening and closing files is very important, as the operating system will lock write access to a file while it is open. A long-running process that locks a file might cause problems for other processes that also need that file. Consequently, the process of opening and closing a file should be accomplished using the with statement, as shown below.

In [1]: with open('qbdata.txt') as f:
   ...:     lines = [line for line in f]
   ...: 

In [2]: lines[:5]
Out[2]: 
['Colt McCoy QB CLE  135 222 1576    6   9   60.8%   74.5\n',
 'Josh Freeman QB TB 291 474 3451    25  6   61.4%   95.9\n',
 'Michael Vick QB PHI    233 372 3018    21  6   62.6%   100.2\n',
 'Matt Schaub QB HOU 365 574 4370    24  12  63.6%   92.0\n',
 'Philip Rivers QB SD    357 541 4710    30  13  66.0%   101.8\n']

Using the with statement when working with files in Python is considered a best-practice, as it guarantees that files are properly opened and closed at the right time.

# open is a context manager, illustrated by the existence of __enter__ and
# __exit__
In [3]: f = open('qbdata.txt')

In [4]: f.__enter__
Out[4]: <function TextIOWrapper.__enter__>

In [5]: f.__exit__
Out[5]: <function TextIOWrapper.__exit__>

6.1.5. Using context managers in comprehensions¶

The with statement provides a useful abstraction for managing resources like file input/output in imperative code.

# Proper imperative context management
In [6]: with open('qbdata.txt') as f:
   ...:     lines = [line for line in f]
   ...: 

In [7]: lines[:5]
Out[7]: 
['Colt McCoy QB CLE  135 222 1576    6   9   60.8%   74.5\n',
 'Josh Freeman QB TB 291 474 3451    25  6   61.4%   95.9\n',
 'Michael Vick QB PHI    233 372 3018    21  6   62.6%   100.2\n',
 'Matt Schaub QB HOU 365 574 4370    24  12  63.6%   92.0\n',
 'Philip Rivers QB SD    357 541 4710    30  13  66.0%   101.8\n']

The more_itertools module provides with_iter, which can be used to embed context managers in expressions as if they were any other iterable object. Similar to the with statement, the with_iter will assure that a resource is managed correctly behind the scenes.

# Proper context management in an expression/comprehension
In [8]: from more_itertools import with_iter

In [9]: lines = [line for line in with_iter(open('qbdata.txt'))]

In [10]: lines[:5]
Out[10]: 
['Colt McCoy QB CLE  135 222 1576    6   9   60.8%   74.5\n',
 'Josh Freeman QB TB 291 474 3451    25  6   61.4%   95.9\n',
 'Michael Vick QB PHI    233 372 3018    21  6   62.6%   100.2\n',
 'Matt Schaub QB HOU 365 574 4370    24  12  63.6%   92.0\n',
 'Philip Rivers QB SD    357 541 4710    30  13  66.0%   101.8\n']

Using with_iter with list comprehension will be our go-to method for reading files. Next, we look at writing data out to files.

# Installing a package with pip.main
In [11]: import pip

In [12]: pip.main(['install', 'more-itertools'])
Collecting more-itertools
  Using cached more_itertools-2.5.0-py3-none-any.whl
Requirement already satisfied (use --upgrade to upgrade): six<2.0.0,>=1.0.0 in /Users/tiverson/.pyenv/versions/3.5.2/envs/runestone/lib/python3.5/site-packages (from more-itertools)
Installing collected packages: more-itertools
Successfully installed more-itertools-2.5.0

6.1.6. Writing Text Files¶

One of the most commonly performed data processing tasks is to read data from a file, manipulate it in some way, and then write the resulting data out to a new data file to be used for other purposes later. To accomplish this, the open function discussed above can also be used to create a new file prepared for writing. Note in Table 1 above that the only difference between opening a file for writing and opening a file for reading is the use of the 'w' flag instead of the 'r' flag as the second parameter. When we open a file for writing, a new, empty file with that name is created and made ready to accept our data. As before, the function returns a reference to the new file object.

Table 2 above shows one additional file method that we have not used thus far. The write method allows us to add data to a text file. Recall that text files contain sequences of characters. We usually think of these character sequences as being the lines of the file where each line ends with the newline \n character. Be very careful to notice that the write method takes one parameter, a string. When invoked, the characters of the string will be added to the end of the file. This means that it is the programmers job to include the newline characters as part of the string if desired.

As an example, consider the qbdata.txt file once again. Assume that we have been asked to provide a file consisting of only the names of the quarterbacks. In addition, the names should be in the order last name followed by first name with the names separated by a comma. This is a very common type of request, usually due to the fact that someone has a program that requires its data input format to be different from what is available.

To construct this file, we will approach the problem using a similar algorithm as above. We start by reading the lines into a list called lines. Then the lines are split into rows using whitespace.

In [11]: from more_itertools import with_iter

In [12]: lines = [line for line in with_iter(open('qbdata.txt'))]

In [13]: split_line = lambda line: line.split()

In [14]: rows = [split_line(line) for line in lines]

In [15]: rows[:2]
Out[15]: 
[['Colt',
  'McCoy',
  'QB',
  'CLE',
  '135',
  '222',
  '1576',
  '6',
  '9',
  '60.8%',
  '74.5'],
 ['Josh',
  'Freeman',
  'QB',
  'TB',
  '291',
  '474',
  '3451',
  '25',
  '6',
  '61.4%',
  '95.9']]

These lines need to be transformed into rows that contain the lastname and first names, in that order and separated by commas. This is accomplished with two comprehensions, first transforming the rows (list of strings) into rows (list of strings) and then joining the rows (list of strings) into a string using the join method with a , as the “glue.”

In [16]: names = [(row[1], row[0]) for row in rows]

In [17]: comma_separate = lambda row: ",".join(row)

In [18]: new_lines = [comma_separate(row) for row in names]

In [19]: new_lines[:5]
Out[19]: ['McCoy,Colt', 'Freeman,Josh', 'Vick,Michael', 'Schaub,Matt', 'Rivers,Philip']

When we run this program, we see the lines of output on the screen. Once we are satisfied that it is creating the appropriate output, the next step is to add the necessary pieces to produce an output file and write the data lines to it. To start, we need to open a new output file by adding another with statement, where outfile will be the alias for open("qbnames.csv",'w'), using the 'w' flag. We can choose any file name we like. If the file does not exist, it will be created. However, if the file does exist, it will be reinitialized as empty and you will lose any previous contents.

Once the file has been created, we just need to call the print function passing the file object in as a parameter. By default, the print function will append a \n to each line. See help(print) for more information about the optional end and sep arguments.

In [20]: with open('qbnames.csv', 'w') as outfile:
   ....:      for line in new_lines:
   ....:           print(line, file = outfile)
   ....: 

The contents of the qbnames.csv file can be shown using the IPython ! magic to make a system call (to bash). In this case we pipe cat filename through the head function to limit the output to the first 5 lines (i.e. -n 5).

In [21]: !cat qbnames.csv | head -n 5
McCoy,Colt
Freeman,Josh
Vick,Michael
Schaub,Matt
Rivers,Philip

6.1.7. Comprehensions and Side Effects¶

Did you notice that the last example was one of the few times that a for loop was used in this book? We could have accomplished the same outcome, writing the data to a file, using a comprehension. Why not use this approach?

Consider the following list comprehension, which mirrors the loop from the last problem.

In [22]: with open('qbnames.csv', 'w') as outfile:
   ....:     out = [print(line, file = outfile) for line in new_lines]
   ....: 

In [23]: out[:10]
Out[23]: [None, None, None, None, None, None, None, None, None, None]

In [24]: !cat qbnames.csv | head -n 5
McCoy,Colt
Freeman,Josh
Vick,Michael
Schaub,Matt
Rivers,Philip

Recall that print is a void function which returns nothing, i.e. None. Consequently, the above expression produces a list of references to None, one for each line that was written to the file. This is meaningless data! The main theme of this text has been referentially transparent expressions, but this last expression is at odds with this theme.

There are two alternatives to the last example that will at least return meaningful data.

6.1.8. Maintaining the sequence with do¶

First, the do function from the toolz.functoolz module takes func and x as inputs, where func is a side-effecting function like print. Calling do will call func with the argument x, then return x (as opposed to the output of func(x)). Using this function in a list comprehension will result in the output of the comprehension will result in the output of the comprehension to be the same as the original input and this output could then be used in later expression. Note that do only works with unary functions, so we will need to create print_to_file to wrap the print(line, file=outfile) call.

In [25]: from toolz.functoolz import do

In [26]: with open('qbnames.csv', 'w') as outfile:
   ....:     print_to_file = lambda line: print(line, file = outfile)
   ....:     out = [do(print_to_file,line) for line in new_lines]
   ....: 

In [27]: out[:10]
Out[27]: 
['McCoy,Colt',
 'Freeman,Josh',
 'Vick,Michael',
 'Schaub,Matt',
 'Rivers,Philip',
 'Hasselbeck,Matt',
 'Clausen,Jimmy',
 'Flacco,Joe',
 'Orton,Kyle',
 'Campbell,Jason']

In [28]: !cat qbnames.csv | head -n 5
McCoy,Colt
Freeman,Josh
Vick,Michael
Schaub,Matt
Rivers,Philip

6.1.9. Abstracting mapping do onto a sequence with side_effect¶

The pattern of applying do on a unary function inside a comprehension can be accomplished using the side_effect function from the more_itertools module, where side_effect(print_to_file, seq) is equivalent to the following comprehension.

[do(print_to_file, line) for line in seq]

It is important to note that side_effect, is a lazy construct that must be forced to completion. On way to accomplish this is to wrap the call in list, which will force completion and return the output. This approach allows this expression to be embedded in other expressions that need this output.

In [29]: from more_itertools import side_effect

In [30]: with open('qbnames.csv', 'w') as outfile:
   ....:     print_to_file = lambda line: print(line, file = outfile)
   ....:     out = list(side_effect(print_to_file, new_lines))
   ....: 

In [31]: out[:5]
Out[31]: ['McCoy,Colt', 'Freeman,Josh', 'Vick,Michael', 'Schaub,Matt', 'Rivers,Philip']

In [32]: !cat qbnames.csv | head -n 5
McCoy,Colt
Freeman,Josh
Vick,Michael
Schaub,Matt
Rivers,Philip

If, on the other hand, we do not need the output of the side_effect call, the explicit method for throwing away this output is to use consume from more_itertools.

In [33]: from more_itertools import side_effect, consume

In [34]: with open('qbnames.csv', 'w') as outfile:
   ....:     print_to_file = lambda line: print(line, file = outfile)
   ....:     consume(side_effect(print_to_file, new_lines))
   ....: 

In [35]: !cat qbnames.csv | head -n 5
McCoy,Colt
Freeman,Josh
Vick,Michael
Schaub,Matt
Rivers,Philip

6.1.10. Alternative File Reading Methods¶

Again, recall the contents of the qbdata.txt file.

Colt McCoy QB CLE  135 222 1576    6   9   60.8%   74.5
Josh Freeman QB TB 291 474 3451    25  6   61.4%   95.9
Michael Vick QB PHI    233 372 3018    21  6   62.6%   100.2
Matt Schaub QB HOU 365 574 4370    24  12  63.6%   92.0
Philip Rivers QB SD    357 541 4710    30  13  66.0%   101.8
Matt Hasselbeck QB SEA 266 444 3001    12  17  59.9%   73.2
Jimmy Clausen QB CAR   157 299 1558    3   9   52.5%   58.4
Joe Flacco QB BAL  306 489 3622    25  10  62.6%   93.6
Kyle Orton QB DEN  293 498 3653    20  9   58.8%   87.5
Jason Campbell QB OAK  194 329 2387    13  8   59.0%   84.5
Peyton Manning QB IND  450 679 4700    33  17  66.3%   91.9
Drew Brees QB NO   448 658 4620    33  22  68.1%   90.9
Matt Ryan QB ATL   357 571 3705    28  9   62.5%   91.0
Matt Cassel QB KC  262 450 3116    27  7   58.2%   93.0
Mark Sanchez QB NYJ    278 507 3291    17  13  54.8%   75.3
Brett Favre QB MIN 217 358 2509    11  19  60.6%   69.9
David Garrard QB JAC   236 366 2734    23  15  64.5%   90.8
Eli Manning QB NYG 339 539 4002    31  25  62.9%   85.3
Carson Palmer QB CIN   362 586 3970    26  20  61.8%   82.4
Alex Smith QB SF   204 342 2370    14  10  59.6%   82.1
Chad Henne QB MIA  301 490 3301    15  19  61.4%   75.4
Tony Romo QB DAL   148 213 1605    11  7   69.5%   94.9
Jay Cutler QB CHI  261 432 3274    23  16  60.4%   86.3
Jon Kitna QB DAL   209 318 2365    16  12  65.7%   88.9
Tom Brady QB NE    324 492 3900    36  4   65.9%   111.0
Ben Roethlisberger QB PIT  240 389 3200    17  5   61.7%   97.0
Kerry Collins QB TEN   160 278 1823    14  8   57.6%   82.2
Derek Anderson QB ARI  169 327 2065    7   10  51.7%   65.9
Ryan Fitzpatrick QB BUF    255 441 3000    23  15  57.8%   81.8
Donovan McNabb QB WAS  275 472 3377    14  15  58.3%   77.1
Kevin Kolb QB PHI  115 189 1197    7   7   60.8%   76.1
Aaron Rodgers QB GB    312 475 3922    28  11  65.7%   101.2
Sam Bradford QB STL    354 590 3512    18  15  60.0%   76.5
Shaun Hill QB DET  257 416 2686    16  12  61.8%   81.3

In addition to using a list comprehension to read in the lines, we can simplify this process using readline. The readlines method returns the contents of the entire file as a list of strings, where each item in the list represents one line of the file. It is also possible to read the entire file into a single string with read. Table 2 summarizes these methods and the following session shows them in action.

Note that we need to reopen the file before each read so that we start from the beginning. Each file has a marker that denotes the current read position in the file. Any time one of the read methods is called the marker is moved to the character immediately following the last character returned. In the case of readline this moves the marker to the first character of the next line in the file. In the case of read or readlines the marker is moved to the end of the file.

In [36]: with  open("qbdata.txt", "r") as infile:
   ....:     linelist = infile.readlines()
   ....: 

In [37]: len(linelist)
Out[37]: 34

In [38]: linelist[0:4]
Out[38]: 
['Colt McCoy QB CLE  135 222 1576    6   9   60.8%   74.5\n',
 'Josh Freeman QB TB 291 474 3451    25  6   61.4%   95.9\n',
 'Michael Vick QB PHI    233 372 3018    21  6   62.6%   100.2\n',
 'Matt Schaub QB HOU 365 574 4370    24  12  63.6%   92.0\n']