Big Data

A Brief Overview of Pandas DataFrames

panda-with-pile-of-bamboo-on-stomach

Significantly more adorable… slightly less helpful for data wrangling

This article is the continuation of my previous article. Here, we will be discussing another datatype, Dataframes.

Dataframes are going to be the main tool that developers use when working with pandas.

You may also like: Understanding NumPy.

Prerequisites

Python’s pandas module should be installed in the system, and, if it is not already, you can install it using:

pip install pandas 

(if you have installed python by directly going to https://www.python.org/downloads/ )

OR

conda install pandas 

(if you have Anaconda distribution of python)

DataFrames and Pandas

DataFrames are the workhorse of pandas and are directly inspired by the R programming language. We can think of a DataFrame as a bunch of Series objects put together to share the same index. Let’s use pandas to explore this topic.

import pandas as pd import numpy as np from numpy.random import randn np.random.seed(101)

To generate some random numbers, we use seed here.

Let’s create a dataframe now:

df = pd.Dataframe

If you are using Jupyter Notebook, press shift+tab after  df = pd.Dataframe , and you will see this:

Result of "Shift+tab" after df initialization

Result of “Shift+tab” after df initialization

Check out the docstring and the initial signature for this DataFrame. We have a data argument, index argument (just like Series), but then we have these additional Columns arguments.

Let’s go ahead and create it with some random data, and we’ll see what a DataFrame actually looks like. For data, we are using randn(5,4) , for index, we are using a list of characters, and for columns, we are using another list of characters.

df = pd.DataFrame(randn(5,4),['A','B','C','D','E'],['W','X','Y','Z']) 

Checking DataFrame content

Checking DataFrame content

So, basically what we have here is a list of columns W, Y, Y, and Z and corresponding rows A, B, C, D, and E. Each of these columns is actually a pandas series, such as W or X or Y or Z, and they all share a common index.

Selection and Indexing:

Let’s grab data from a DataFrame.

Selecting columns:

df['W']

 

Outputting the "W" column to the screen

Outputting the “W” column to the screen

You can check the type using:

type(df['W']) 

This will give pandas.core.series.Series as a result.

You can also check:

type(df) 

which will give pandas.core.frame.DataFrame as a result.

If you want to select multiple columns use:

df[['W','Z']] 

Indexing columns "W" and "Z"

Indexing columns “W” and “Z”

Creating New Columns

df['new'] = df['W'] + df['Y'] 

Creating a new column, "new"

Creating a new column, “new”

Removing Columns

For removing columns, you can just do:

df.drop('new',axis=1) 

Dropping "new" from the DataFrame

Dropping “new” from the DataFrame

Here, you can use shift + tab to check what axis actually refers to. Axis = 0, which is by default is for rows, whereas, Axis = 1 refers to columns. So, here we use axis=1, because we wanted to drop a column.

Note: “new” column still exists; you have to use the inplace parameter to retain this change. Pandas does this so users don’t accidentally lose information. So, use inplace = True.

Using inplace parameter

Using inplace parameter

We can also use  df.drop('E',axis=0)  to drop a row. Try it yourself.

A Quick Question:

Why are the rows 0 and why are the columns 1?

The reference actually comes back to numpy. DataFrames are essentially index markers on top of a numpy array. Using df.shape() results in a tuple (5, 4). For a two-dimensional matrix, at the 0 index are the number of rows (A,B,C,D,E) and then on the index 1 are columns (W,X,Y,Z). This is why rows are referred to as the 0 axis and columns are referred to as 1 axis because it’s directly taken from the shape, just like in a numpy array.

Selecting Rows

There are two ways to select rows in a DataFrame, and you have to call a method for this.

            Select based on label:

             df.loc['A'] 

OR

            Select based on position:

             df.iloc[2] 

Using loc and iloc to select rows

Using loc and iloc to select rows

Note: Not only are all the columns series but the rows are series as well.

Selecting Subsets of Rows and Columns:

For this, you can use:

df.loc[['A','B'],['W','Y']] 

For selecting a particular value, use:

 df.loc['B','Y'] 

Selecting subsets of rows using loc

Selecting subsets of rows using loc

Conditional Selection

A very important feature of pandas is the ability to perform conditional selection using bracket notation. This is going to be very similar to numpy.

Let’s use a comparison operator:

 df > 0 

The result is a DataFrame with boolean values. This returns true if the DataFrame value at that position is greater than zero and false if it is not greater than zero. See below:

Creating masks for our DataFrame

Creating masks for our DataFrame

 df[df>0] 

As you can see, wherever the value is negative, not satisfying the condition, NaN has been returned.

Now, what’s important is, instead of returning NaN, we will return only the rows or columns of a subset of the DataFrame where the conditions are true.

Using masks on the whole DataFrame and one column

Using masks on the whole DataFrame and one column

Let’s say we want to grab information in DataFrame where the column value is 

W>0

, and we want to extract the “Y” column. We can also select a set of columns such as “Y” and “X,” after applying the condition. See below:

Image title

Using multiple Conditions:

For more than one condition, we can use | or &. Remember that we cannot use python’s and/or here.

df[(df['W']>0) & (df['Y'] > 1)] 

Masks with multiple conditions

Masks with multiple conditions

Resetting Indexes

In order to reset the index back to the default which is 1234….n, we use the method reset_index(). We will get the index, reset to a column and the actual index converted to a numerical. But it will not retain the change if you don’t use inplace=True. Pandas uses this inplace argument in many areas, just shift+tab (if using Jupyter Notebook), and you will see it.

df.reset_index() 

Resetting indexes

Resetting indexes

Setting New Index

For setting a new index, first, we have to create a new index. We are using the split() method of a string, which is just a common method for splitting off a blank space. It’s a quick way to create a list 😉

newind = 'WB MP KA TN UP'.split() 

Now, put this list as a column of the dataframe.

df['States'] = newind df 

If we want to use this State column as the index, we should use:

df.set_index('States') 

Result of setting "States" as the new index

Result of setting “States” as the new index

Note: Unless we retain this information of the index, it will overwrite the old index, and we won’t actually be able to retain this information as a new column — unlike resets index that allows us to have that new column.

So, that’s set index versus reset index. 🙂

Here, inplace=True also plays an important role.

I hope, you have enjoyed reading about DataFrames thus far. There’s more to cover in the upcoming article on DataFrames.

Happy learning 🙂

Related Articles

Comment here

This site uses Akismet to reduce spam. Learn how your comment data is processed.