The prerequisite for doing any data-related operations in Python, such as data cleansing, data aggregation, data transformation, and data visualisation, is to load data into Python. Depends on the types of data files (e.g.
.json, Excel spreadsheets, relational databases etc.) and their size, different methods should be applied to deal with this initial operation accordingly. In this post, I will list some common methods for importing data in Python.
As usual, all required data and notebook can be accessed from my Github.
- Python build-in functions (
In general, a text file (
.txt) is the most common file we will deal with. Text files are structured as a sequence of lines, where each line includes a sequence of characters. Let’s assume we need to import in Python the following text file (
Country/Region Mainland China Japan Singapore Hong Kong Japan Thailand South Korea Malaysia Taiwan Germany Vietnam France Macau UK United Arab Emirates US Australia
To import its content to Python, we need to first open it. This step is just like double-clicking a file to open it in our computer system. However, in Python, this is done by calling the
open() built-in function.
open() has a required argument that is the path to the file and an optional argument to indicate the mode (i.e. default argument ‘r’: open for reading; ‘w’: open for writing). With those set,
open() then returns us a file object.
There are three methods to read content (i.e.
readlines()) that can be called on this file object either by one or in combination.
read(size=-1): this reads from the file based on the number of
sizebytes. If no argument is passed or
-1is passed, then the entire file is read.
readline(size=-1): this reads the entire line if no arguments are passed or
-1is passed (Figure 2 upper panel). Or if passed with
size, this reads the
sizenumber of characters from the line. Moreover, multiple
readline()functions (Figure 2 lower panel) can be called sequentially, in which the next
readline()function will continue from the end position of last
readline()function. Note that output of the third
readline()appends an extra newline character(\n, displayed as a new line). This can be avoided by using
readlines(): this reads all the lines or remaining lines from the file object and returns them as a list (Figure 3).
You might notice that all the above codes have
with statements. The
with statement provides a way for ensuring that the file is always closed after open. Without the
with statement, we need explicitly to call
close() for the file object. For instance:
file = open("sample.txt") data = file.read() print(data) file.close()
As it is very easy to forget to close the file, we should always use
with statement. This also provides better syntax and exceptions handling.
sample.txt we just processed had only one field per line, which make it handy to process using just build-in function (
readlines()). However, more frequently we will work with a file that has multiple fields on each line (aka tabular data), as shown in Figure 4.
As we can see that every field on each line is comma-separated, indicating where one field ends and the next field starts. We call this type of file the delimited file. These files are often either comma-separated (
.csv)or tab-separated (
.txt). In rare cases, you may also encounter other delimiters like colon (
:), semi-colon (
;), and characters.
Even though the build-in function can still process these files, it is highly likely to mess up, especially when there are hundreds of fileds per line in some cases. Alternatively, we can use Python’s
csv library designed to read delimited files. Here let’s learn two common functions from this module.
csv.reader(): this reads all lines in the given file and returns a reader object. Then each line can be returned as a list of strings.
csv.DictReader(): if the file has headers (normally the first row that identifies each filed of data), this function reads each line as a
dictwith the headers as
keys(Figure 6, upper panel). We then can access data of each column by calling its fieldname (Figure 6, lower panel)
3. Import data using Pandas
Another very popular option in importing data in Python must go to Pandas, especially when the data size is big (like several hundred MBs). We won’t delve into the specifics of how
pandas works or how to use it. There are many excellent tutorials and books (e.g. Python for Data Analysis, by Wes McKinney, creator of pandas). Here I just show some of the power of
pandas in reading
pd.read_csv(): this reads a
DataFrameobject. An important point here is that pandas is smart enough to automatically tell the header row and data type of each field, which make the downstream analyse more efficient.
pd.read_excel(): this reads an excel file (
.odffile extensions) into a pandas
DataFrame. By default, it only import the first data sheet within the excel file (
sample.xlsxhas multiple sheets), as shown in Figure 8.
To be able to access a specific sheet within the excel file, we can first import the whole excel file using
pd.ExcelFile() and then specify the sheet name when calling
pd.read_excel() (Figure 8).
Moreover, to be handier in accessing all sheets, instead of calling
pd.read_excel() multiple times, we can store all sheets as
dataframe objects inside a
dict (Figure 9). A practical example can also be found in this post.
4. Options for importing large size data
In the age of big data, sometimes, we need to import files from a client or colleague, which may be too large (gigabytes or terabytes) to load into memory. So what should we do to tackle this bottleneck?
Fortunately, Pandas provides
chunksize option to work this around. Essentially, we are not importing the whole file in one go instead of importing partial contents.
In addition, I found a very useful post by Mihail Yanchev, where he provided multiple methods and compared their efficiency in handling this situation. Here I just list those methods mentioned in his post and you can read his post if that is what you are looking for.
- dask.dataframe(): a large parallel DataFrame composed of many smaller Pandas DataFrames, split along the index. A good thing is that most functions used with pandas can also be used with dask.
- datatable: a Python package for manipulating big 2-dimensional tabular data structures (aka data frames, up to 100GB).
Alright, now we know the basics of importing data in Python. Of course, there are many other cases in importing data in Python that I cannot list them all here, like parsing
html using BeautifulSoup and reading sql table using
pd.read_sql_query(). I hope this post gets you to start from the ground and up to explore more possibilities