Pandas is one of the most popular Python libraries for Data Science and Analytics. I like to say it’s the “SQL of Python.” Why? Because pandas helps you to manage two-dimensional data tables in Python. Of course, it has many more features. In this pandas tutorial series, I’ll show you the most important (that is, the most often used) things that you have to know as an Analyst or a Data Scientist. This is the first episode and we will start from the basics!
Note 1: this is a hands-on tutorial, so I recommend doing the coding part with me!
Before we start
If you haven’t done so yet, I recommend going through these articles first:
- How to install Python, R, SQL and bash to practice data science
- Python for Data Science – Basics #1 – Variables and basic operations
- Python Import Statement and the Most Important Built-in Modules
- Top 5 Python Libraries and Packages for Data Scientists
To follow this pandas tutorial…
- You will need a fully functioning data server with Python3, numpy and pandas on it.
Note 1 : Again, with this tutorial you can set up your data server and Python3. And with this article you can set up numpy and pandas, too.
Note 2: or take this step-by-step data server set up video course.
- Next step: log in to your server and fire up Jupyter. Then open a new Jupyter Notebook in your favorite browser. (If you don’t know how to do that, I really do recommend going through the articles I linked in the “Before we start” section.)
Note: I’ll also rename my Jupyter Notebook to “pandas_tutorial_1”.
- Import numpy and pandas to your Jupyter Notebook by running these two lines in a cell:
import numpy as np import pandas as pd
Note: It’s conventional to refer to ‘pandas’ as ‘pd’. When you add the
as pdat the end of your import statement, your Jupyter Notebook understands that from this point on every time you type
pd, you are actually referring to the
Okay, now we have everything! Let’s start with this pandas tutorial!
The first question is:
How to open data files in pandas
You might have your data in .csv files or SQL tables. Maybe Excel files. Or .tsv files. Or something else. But the goal is the same in all cases. If you want to analyze that data using pandas, the first step will be to read it into a data structure that’s compatible with pandas.
Pandas data structures
There are two types of data structures in pandas: Series and DataFrames.
Series: a pandas Series is a one dimensional data structure (“a one dimensional ndarray”) that can store values — and for every value it holds a unique index, too.
DataFrame: a pandas DataFrame is a two (or more) dimensional data structure – basically a table with rows and columns. The columns have names and the rows have indexes.
In this pandas tutorial, I’ll focus mostly on DataFrames. The reason is simple: most of the analytical methods I will talk about will make more sense in a 2D datatable than in a 1D array.
Loading a .csv file into a pandas DataFrame
Okay, time to put things into practice! Let’s load a .csv data file into pandas!
There is a function for it, called
Start with a simple demo data set, called zoo! This time – for the sake of practicing – you will create a .csv file for yourself! Here’s the raw data:
animal,uniq_id,water_need elephant,1001,500 elephant,1002,600 elephant,1003,550 tiger,1004,300 tiger,1005,320 tiger,1006,330 tiger,1007,290 tiger,1008,310 zebra,1009,200 zebra,1010,220 zebra,1011,240 zebra,1012,230 zebra,1013,220 zebra,1014,100 zebra,1015,80 lion,1016,420 lion,1017,600 lion,1018,500 lion,1019,390 kangaroo,1020,410 kangaroo,1021,430 kangaroo,1022,410
Go back to your Jupyter Home tab and create a new text file…
…then copy-paste the above zoo data into this text file…
… and then rename this text file to zoo.csv!
Okay, this is our .csv file.
Now, go back to your Jupyter Notebook (that I named ‘pandas_tutorial_1’) and open this freshly created .csv file in it!
Again, the function that you have to use is:
Type this to a new cell:
pd.read_csv('zoo.csv', delimiter = ',')
And there you go! This is the zoo.csv data file, brought to pandas. This nice 2D table? Well, this is a pandas dataframe. The numbers on the left are the indexes. And the column names on the top are picked up from the first row of our zoo.csv file.
To be honest, though, you will probably never create a .csv data file for yourself, like we just did… you will use pre-existing data files. So you have to learn how to download .csv files to your server!
If you are here from the Junior Data Scientist’s First Month video course then you have already dealt with downloading your .txt or .csv data files to your data server, so you must be pretty proficient in it… But if you are not here from the course (or if you want to learn another way to download a .csv file to your server and to get another exciting dataset), follow these steps:
I’ve uploaded a small sample dataset here: DATASET
If you click the link, the data file will be downloaded to your computer. But you don’t want to download this data file to your computer, right? You want to download it to your server and then load it to your Jupyter Notebook. It only takes two steps.
STEP 1) Go back to your Jupyter Notebook and type this command:
This downloaded the pandas_tutorial_read.csv file to your server. Just check it out:
See? It’s there.
If you click it…
…you can even check out the data in it.
STEP 2) Now, go back again to your Jupyter Notebook and use the same
read_csv function that we have used before (but don’t forget to change the file name and the delimiter value):
The data is loaded into pandas!
Does something feel off? Yes, this time we didn’t have a header in our csv file, so we have to set it up manually! Add the names parameter to your function!
pd.read_csv('pandas_tutorial_read.csv', delimiter=';', names = ['my_datetime', 'event', 'country', 'user_id', 'source', 'topic'])
And with that, we finally loaded our .csv data into a pandas dataframe!
Note 1: Just so you know, there is an alternative method. (I don’t prefer it though.) You can load the .csv data using the URL directly. In this case the data won’t be downloaded to your data server.
Note 2: If you are wondering what’s in this data set – this is the data log of a travel blog. This is a log of one day only (if you are a JDS course participant, you will get much more of this data set on the last week of the course ;-)). I guess the names of the columns are fairly self-explanatory.
Selecting data from a dataframe in pandas
This is the first episode of this pandas tutorial series, so let’s start with a few very basic data selection methods – and in the next episodes we will go deeper!
1) Print the whole dataframe
The most basic method is to print your whole data frame to your screen. Of course, you don’t have to run the
pd.read_csv() function again and again and again. Just store its output the first time you run it!
article_read = pd.read_csv('pandas_tutorial_read.csv', delimiter=';', names = ['my_datetime', 'event', 'country', 'user_id', 'source', 'topic'])
After that, you can call this
article_read value anytime to print your DataFrame!
2) Print a sample of your dataframe
Sometimes, it’s handy not to print the whole dataframe and flood your screen with data. When a few lines is enough, you can print only the first 5 lines – by typing:
Or the last few lines by typing:
Or a few random lines by typing:
3) Select specific columns of your dataframe
This one is a bit tricky! Let’s say you want to print the ‘country’ and the ‘user_id’ columns only.
You should use this syntax:
Any guesses why we have to use double bracket frames? It seems a bit over-complicated, I admit, but maybe this will help you remember: the outer bracket frames tell pandas that you want to select columns, and the inner brackets are for the list (remember? Python lists go between bracket frames) of the column names.
By the way, if you change the order of the column names, the order of the returned columns will change, too:
This is the DataFrame of your selected columns.
Note: Sometimes (especially in predictive analytics projects), you want to get Series objects instead of DataFrames. You can get a Series using any of these two syntaxes (and selecting only one column):
4) Filter for specific values in your dataframe
If the previous one was a bit tricky, this one will be really tricky!
Let’s say, you want to see a list of only the users who came from the ‘SEO’ source. In this case you have to filter for the ‘SEO’ value in the ‘source’ column:
article_read[article_read.source == 'SEO']
It’s worth it to understand how pandas thinks about data filtering:
STEP 1) First, between the bracket frames it evaluates every line: is the
article_read.source column’s value
'SEO' or not? The results are boolean values (
STEP 2) Then from the
article_read table, it prints every row where this value is
True and doesn’t print any row where it’s
Does it look over-complicated? Maybe. But this is the way it is, so let’s just learn it because you will use this a lot! 😉
Functions can be used after each other
It’s very important to understand that pandas’s logic is very linear (compared to SQL, for instance). So if you apply a function, you can always apply another one on it. In this case, the input of the latter function will always be the output of the previous function.
E.g. combine these two selection methods:
This line first selects the first 5 rows of our data set. And then it takes only the ‘country’ and the ‘user_id’ columns.
Could you get the same result with a different chain of functions? Of course you can:
In this version, you select the columns first, then take the first five rows. The result is the same – the order of the functions (and the execution) is different.
One more thing. What happens if you replace the ‘article_read’ value with the original
pd.read_csv('pandas_tutorial_read.csv', delimiter=';', names = ['my_datetime', 'event', 'country', 'user_id', 'source', 'topic'])[['country', 'user_id']].head()
This will work, too – only it’s ugly (and inefficient). But it’s really important that you understand that working with pandas is nothing but applying the right functions and methods, one by one.
As always, here’s a short assignment to test yourself! Solve it, so the content of this article can sink in better!
Select the user_id, the country and the topic columns for the users who are from country_2! Print the first five rows only!
Okay, go ahead and solve it!
And here’s my solution!
It can be a one-liner:
article_read[article_read.country == 'country_2'][['user_id','topic', 'country']].head()
Or, to be more transparent, you can break this into more lines:
ar_filtered = article_read[article_read.country == 'country_2'] ar_filtered_cols = ar_filtered[['user_id','topic', 'country']] ar_filtered_cols.head()
Either way, the logic is the same. First you take your original dataframe (
article_read), then you filter for the rows where the country value is country_2 (
[article_read.country == 'country_2']), then you take the three columns that were required (
[['user_id','topic', 'country']]) and eventually you take the first five rows only (
You are done with the first episode of my pandas tutorial series! Great job! In the next article, you can learn more about the different aggregation methods (e.g. sum, mean, max, min) and about grouping (so basically about segmentation). Stay with me: Pandas Tutorial, Episode 2!
- If you want to learn more about how to become a data scientist, take my 50-minute video course: How to Become a Data Scientist. (It’s free!)
- Also check out my 6-week online course: The Junior Data Scientist’s First Month video course.