Introduction
Learning data science is an exciting and rewarding journey. Part of this journey is learning the tools of a data scientist. One of the challenges when starting to learn data science is knowing what tools you should be using and learning.
Python is an excellent choice for working as a data scientist - partly because of the rich ecosystem of Python libraries. We suggest starting your data science journey with two Python libraries - pandas and matplotib:
- pandas for loading, transforming and saving data.
- matplotib for visualizing data.
Together these two libraries offer enough to start doing meaningful data work in Python.
pandas
pandas is a Python library for working with tabular data - data with rows and columns. If you have worked with Excel, you have already worked with tabular data.
In this article we will work with a simple cities dataset - a tabular dataset with three rows and three columns:
Installing and using pandas
You can install pandas with pip, a Python package manager:
$ pip install pandasYou can also find a notebook with all the code we develop in this article on Github or on Google Colab.
Pandas is used by importing the pandas package, commonly aliased to pd:
import pandas as pdWhat is a DataFrame?
At the core of pandas is the DataFrame (a pd.DataFrame Python object), composed of three parts:
- index - 1-dimensional row labels,
- columns - 1-dimensional column labels,
- values - 2-dimensional data.
Both the index and columns are labels - they tell us what a row or column represents, such as the population column label.
Creating a DataFrame
One way to create a DataFrame is from a Python dictionary, with:
- the column names as dictionary keys,
- the data as the dictionary values.
Below we create a cities DataFrame from a dictionary:
import pandas as pd
pd.DataFrame({
'city': ['auckland', 'berlin', 'london'],
'population': [1.6, 3.6, 8.9],
'hemisphere': ['south', 'north', 'north']
})
"""
city population hemisphere
0 auckland 1.6 south
1 berlin 3.6 north
2 london 8.9 north
"""Loading CSV data into a DataFrame
Being able to construct DataFrames from Python objects such as dictionaries and lists is useful for constructing data for unit tests.
A more common way to create a DataFrame is from a file - commonly a CSV file. This allows us to work with data saved on files on our local machine.
Pandas handles this with pd.read_csv, which reads data from a CSV file on our local computer. This same function can also be used to read data from the Internet.
Below we read in our cities dataset using pd.read_csv to read from a Github URI:
import pandas as pd
data = pd.read_csv(
'https://raw.githubusercontent.com/ADGEfficiency/data-science-south-data/main/cities/cities.csv'
)
"""
city population hemisphere
0 auckland 1.6 south
1 berlin 3.6 north
2 london 8.9 north
"""This is the way we will load our cities dataset in rest of this post - reading directly from a public URI (such as a URL).
Index, Columns & Values
We can access the three parts of our DataFrame as attributes of an initialized pd.Dataframe object:
data.index
# RangeIndex(start=0, stop=3, step=1)
data.columns
# Index(['city', 'population', 'hemisphere'], dtype='object')
data.values
"""
[['auckland' 1.6 'south']
['berlin' 3.6 'north']
['london' 8.9 'north']]
"""The index pandas created for us is just a range index (also called an integer index) - rows are labelled with a sequence of integers ([0, 1, 2]):
data.index
# RangeIndex(start=0, stop=3, step=1)
data
"""
city population hemisphere
0 auckland 1.6 south
1 berlin 3.6 north
2 london 8.9 north
"""We can replace this with more meaningful index using set_index, turning the city column into the index:
data.set_index('city')
"""
population hemisphere
city
auckland 1.6 south
berlin 3.6 north
london 8.9 north
"""We have lost our original integer index of [0, 1, 2] and gained an index of city - great!
Selecting rows and columns in pandas
A basic operation in data analysis is selection - to select rows and columns. There are two ways to do this in pandas - loc and iloc.
Why do we need two ways to select?
We want to select rows or columns in two ways:
- a label using
loc, - a position using
iloc.
Both require us to specify both the row and columns (by either label or position) - using : to select the entire row or column.
loc uses labels
loc selects based on the label of the row and column.
loc allows us to use the labels of the index and columns - we use it to select data based on labels:
# select the berlin row, all columns
data.loc['berlin', :]
# select all rows, second colun
data.loc[:, 'population']iloc uses position
iloc selects based on the integer position of the row and column.
iloc allows us to use the position of rows and columns - we use it to select data based on position:
# select the first row, all columns
data.iloc[0, :]
# select all rows, second column
data.iloc[:, 1]Selecting based on position is very useful when your data is sorted.
iloc is why a range index isn’t that useful in pandas - we can always use iloc to select data based on it’s position.
Now we have been introduced to loc & iloc, let’s use them to answer two questions about our cities dataset.
What is the population of Auckland?
We can answer this using loc, selecting the auckland row and the population column:
data.loc['auckland', 'population']
# 1.6Which hemisphere is our first city in?
We can answer this using iloc to select the first row with 0 and loc to select the population column:
data.iloc[0].loc['hemisphere']
# northFiltering with boolean masks in pandas
Another basic operation in data analysis is filtering - selecting rows or columns based on conditional logic (if statements, equalities like == and inequalities like > or <).
We can filter in pandas with a boolean mask, which can be created with a conditional statement:
import pandas as pd
data = pd.read_csv('https://raw.githubusercontent.com/ADGEfficiency/data-science-south-data/main/cities/cities.csv')
# create our boolean mask
# with the conditional 'population < 2.0'
# aka population less than 2.0
mask = data.loc[:, 'population'] < 2.0
"""
city
auckland True
berlin False
london False
Name: population, dtype: bool
"""The boolean mask is an array of either True or False - here indicating True if the population of the city is less than 2.
We can use our boolean mask to filter our dataset with loc - loc understands how to use a boolean mask:
subset = data.loc[mask, :]
"""
population hemisphere
city
auckland 1.6 south
"""Why do we need boolean masks if we can select with loc or iloc?
The power of using boolean masks is that we can select many rows at once.
Below we create a boolean mask based on the hemisphere of the city - then using this mask to select two rows:
mask = data.loc[:, 'hemisphere'] == 'north'
"""
city
auckland False
berlin True
london True
Name: hemisphere, dtype: bool
"""
subset = data.loc[mask, :]
"""
population hemisphere
city
berlin 3.6 north
london 8.9 north
"""Aggregation with groupby in pandas
The final data analysis operation we will look at is aggregation.
In pandas aggregation can be done by grouping - using the groupby method on a DataFrame.
The two step groupby workflow
In pandas aggregation happens in two steps:
- creating groups based on a column, such as by
hemisphere, - applying an aggregation function to each group, such as counting or averaging.
Aggregation allows us to estimate statistics, lets use it to answer a few more questions.
What is the average population in each hemisphere?
We can answer this with our two step workflow:
groupby('hemisphere')to group by hemisphere,mean()to calculate the average for each of our hemisphere groups (northandsouth).
import pandas as pd
data = pd.read_csv('https://raw.githubusercontent.com/ADGEfficiency/data-science-south-data/main/cities/cities.csv').set_index('city')
data.groupby('hemisphere').mean()
"""
population
hemisphere
north 6.25
south 1.60
"""What is the total population in each hemisphere?
Our two steps to answer this question are:
- grouping by
hemisphere, - aggregating with a
sum.
data.loc[:, ['population', 'hemisphere']].groupby('hemisphere').sum()
"""
population
hemisphere
north 12.5
south 1.6
"""Saving our data to CSV
The last step of our process is going to be to save our data to our harddrive, in Excel friendly CSV format:
data.to_csv('groups.csv')Full pandas code
That is it for our first look at pandas - all the code we looked at above is given below in full:
import pandas as pd
# read dataset from github url
data = pd.read_csv('https://raw.githubusercontent.com/ADGEfficiency/data-science-south-data/main/cities/cities.csv').set_index('city')
# save cities dataset to local machine
data.to_csv('cities.csv')
# select the berlin row, all columns
data.loc['berlin', :]
# select all rows, second colun
data.loc[:, 'population']
# select the first row, all columns
data.iloc[0, :]
# select all rows, second column
data.iloc[:, 1]
# What is the population of Auckland?
data.loc['auckland', 'population']
# Which hemisphere is our first city in?
data.iloc[0].loc['hemisphere']
# select population less than 2
mask = data.loc[:, 'population'] < 2.0
subset = data.loc[mask, :]
# northern hemisphere countries
mask = data.loc[:, 'hemisphere'] == 'north'
subset = data.loc[mask, :]
# average population in each hemisphere
data.groupby('hemisphere').mean()
# total population in each hemisphere
data.loc[:, ['population', 'hemisphere']].groupby('hemisphere').sum()
# save to csv file on local computer
data.to_csv('groups.csv')matplotlib
matplotlib is a Python library for creating visualizations of data.
One of the challenges with learning and using matplotlib is that it offers multiple ways to plot data. Mastering matplotlib requires understanding which API is best for your data and workflow.
We will use a single matplotlib workflow - the one we use the most ourselves. It offers the flexibility to plot multiple charts in the same figure, and integrates nicely with pandas.
Installing and using matplotlib
You can install matplotlib with pip, a Python package manager:
$ pip install matplotlibYou can also find a notebook with all the code on Github or on Google Colab.
Matplotlib is used by importing the matplotlib package, commonly aliasing the module pyplot as plt:
import matplotlib.pyplot as pltWhat is a matplotlib figure made of?
The core components in matplotlib are the figure and axes.
One figure can have many axes:
A common point of confusion is between axis (such as the x or y axis) and a matplotlib axes - they are not the same thing!
One figure can have multiple axes - each axes is a separate plot or chart. Each of these axes has it’s own x-axis and y-axis (one of each).
We can create these two objects using plt.subplots - creating one figure with two axes:
import matplotlib.pyplot as plt
fig, axes = plt.subplots(ncols=2)All we’ve done above is create a figure with two axes - both empty.
Next we load our data with pandas, and create a plot on our first axes:
import pandas as pd
# run a simple data pipeline - load data from CSV with pandas
data = pd.read_csv('https://raw.githubusercontent.com/ADGEfficiency/data-science-south-data/main/cities/cities.csv').set_index('city')
# access first axes and plot a line
data.plot('population', ax=axes[0], kind='bar')Automatically makes labels for the x and y axis - quite nice!
Our second axes is still empty - we can plot something on it by passing ax=axes[1] into another plot call on our DataFrame:
# access first axes and plot a scatter plot
data.plot('land-area', 'population', ax=axes[1], kind='scatter')
Now we can see two visualizations of our data - a bar chart and a scatter plot.
The final step in our pipeline is to save our figure to a PNG file on our local machine:
fig.savefig('cities.png')Full matplotlib code
The full code for our visualization pipeline is below:
import matplotlib.pyplot as plt
import pandas as pd
# create one figure with two axes
fig, axes = plt.subplots(nrows=2)
# run a simple data pipeline
data = pd.read_csv('https://raw.githubusercontent.com/ADGEfficiency/data-science-south-data/main/cities/cities.csv').set_index('city')
# access first axes and plot a line
data.plot(y='population', ax=axes[0], kind='bar')
# access first axes and plot a scatter plot
data.plot('land-area', 'population', ax=axes[1], kind='scatter')
# small trick to get x-axis labels to play nice
plt.tight_layout()
# save the figure as a png file
fig.savefig('cities.png')Summary
Thanks for reading!
There is much more to learn with both of these libraries.
Suggested next steps for pandas are:
- learn about
pd.Series- another pandas object used to store data, - learn how to save data to Parquet format (and why),
- see how we can use
.aggfor different aggregations with a single groupby.
Suggested next steps for matplotlib are:
- learn to plot data using
plt.plot, - learn to plot data using
ax.plot, - check out seaborn - a library that extends matplotlib with charts like the pairplot.