Python for Data Science

By | June 13, 2020
Python for Data Science

What is Data Science?

Termed as an interdisciplinary field, Data science is truly a field that consists of collection, preparation, analysis, visualization, and reporting of data to get useful insights from the same and make future predictions. For example, a company whose sales are going down can collect data from the time when the sales dropped and analyze the data to find the root cause, patterns, suggest future courses and make a forecast based on the past and current conditions.

Why Python for data science?

To perform a detailed analysis of data is not easy – by both humans and computers. Computers can, however, do it with ease when the inputs are given properly. But we can’t expect the input to be perfect always as it can come from various sources and is raw and unstructured.


There begins the journey of Python for data science!

At every stage of data science, Python is used to perform various tasks. Python has a rich set of libraries and functions that are extremely useful for data science.

Using python, you can load the data, clean, prepare and transform data, analyze the same using machine learning algorithms, and generate reports and visualizations.

Python is simple to learn, has a syntax that’s just like plain English, and most importantly has an exhaustive set of libraries that cater to all data science needs. All you need to know is how to use different functions for different data science tasks.

Python has handy help and supports documentation.

The online community is vast and helpful.

Further, the Python package itself comes with example datasets for you to play around with and learn at the same time.

A little bit of Python…

Let us take a quick tour of Python. We will be focusing on the concepts that are most useful for data science throughout this article, but first we need to get the basics right.

How to learn Python for data science

Installation and basic syntax

Let us start with installing Python, then we will get onto simple programs where we explore variables, data types, loops, conditions, operators, functions, and more.

Python installation is very simple. Just download the distribution (Windows, Linux, Mac) from Python official downloads page.

Once you download and install as per the instructions, open the command prompt (terminal) and type python. You will get the version number. You can then explore and check out various commands.

You can directly work on the command line, however, for a more interactive experience, you can install Jupyter or pycharm. To install jupyter, use the command “install jupyter”. To run jupyter just type “jupyter notebook” and it will open the notebook in your browser. You can select a new document and start coding.

Our first Python program

If you haven’t done your set up – go for it. There is no better time than now. Let us write a simple program that scans a few words from a list to find a particular word. Don’t worry about the explanations yet, we will go into that later.

mylist = ['apple', 'grapes', 'banana', 'orange', 'mango']
count = len(mylist)
# iterate through each item in the list
for i in mylist:
# if it is banana
if(i == 'banana'):
print("i dont want to eat bananas")
#if it is mango, a special message
elif(i == 'mango'):
print("i love mango")
# common message for others
print("good fruit : " + i)
#print the fruit list

This is a simple program that takes a list of fruits, iterates over each item and prints certain messages.

Python syntax

Python is very particular about indentation. If you leave or omit unnecessary spaces, the code will not work. For example,

if num > 10:
print(“You win”)
is not treated the same as –
if num > 10:
print(“You win”)

The latter will give an error because of the indentation.

Python comments

Python supports one-line comments using ‘#’.

#This is a single-line comment.

If your comments occupy more than one line, you can use “”” (triple quotes).

This is a multi-line
Comment. Of course, not as
neat as a single line.


There is no need to declare variables in Python unlike other languages like C++, Java, etc… You can just assign a value while creation itself.

x = 44
name = “TechGeeks” or name = ‘TechGeeks’

Notice that there is no need to specify the type as well. However, variables should start with a letter or underscore, but never with a number. Variable names are case-sensitive and can contain alpha-numeric characters and underscores.

You can assign values to multiple variables in a single line as,

X, y, z = 10, 20, 30

To make a variable global, we can add the keyword global to the variable.

Data Types

Python supports a lot of built-in data types –

Numeric int, float, complex
Text String (str)
Boolean bool
Sequence list, range, tuple
Set set, frozenset
Map dictionary (dict)
Binary bytes, bytearray, memoryview

Since we do not explicitly declare variables, to get the data type of a variable, we can use the method type(). For example, type(y) where y is any variable.

It is also possible to specify the data type using the constructor methods :-

C = int(10)
Mylist = list([1, 2, 3, 4, 5])

Some examples of each data type –

Numbers a = 10
f = 12.3455
b = 2e4
cplx = 1 + 2j
Strings name = ‘tgb’
name = “tgb”There are many string operations like slicing, negative index, length, lower(), upper(), split(), replace() etc…Strings can be concatenated using + operator
Booleans statements like 1 > 2, 5> 4, etc… will return a boolean false or true.

Booleans are also used for evaluating if conditions.

Operators there are many operators like + , -, *, /, **, =, +=, -=, >>=, ==, >, >=, and, or, not, is, is not, in, not in etc…

x + y

a not in mylist

y > 5
a !=b

Lists mylist = [“machine learning”, “data science”, “artificial intelligence”]

intlist = [1, 3, 5 ,7, 9]

Tuples tuple is unchangeable and ordered

mytuple = (“machine learning”, “data science”, “artificial intelligence”)

Sets unordered and unindexed collection

myset = {“machine learning”, “data science”, “artificial intelligence”}

mydict = {
"0098": "maths",
"0101": "english",
"076": "science"

Conditional statement (if-else)

a, b = 10, 12
print(“a wins”)
print(“a & b win”)
print(“b wins”)

While loop

i = 1

For loop

For and if are the most used statements in Python for data science. ‘for’ is used for iteration.

mylist = [‘maths’, ‘english’, ‘science’]
for i in mylist:

This will print the contents of the list one by one.


Just like other languages, we can define functions that can have input and output parameters and are executed when called.

def myadd(a, b, c):
return a+b+c

To call the function, just type it with the parameters as myadd(1, 2, 3)

Python libraries for data science

There are many useful libraries in Python for data science –

1. Numpy

Numpy means Numerical (Num) Python (Py). This library has rich features like basic linear algebra, n-dimensional array, random number capabilities, matrix computations, fourier transforms, integration tools for C, C++, and Fortran. The most powerful feature of numpy used in all the data science projects is the n-dimensional array. As Python itself supports only lists (and not arrays), numpy can be used to convert the list to an array. This helps in the data evaluation stage of data science. Numpy can be imported as –

import numpy as np

Before using numpy, you should install it by using

pip install numpy

Here is a snippet of how n-d-arrays can be created using numpy –

import numpy as np
#simple array
mylist = [1,2,3,4]
arr = np.array(mylist)
arr[3] = 7
#create an array of zeroes
arrzero = np.zeros([3, 3], dtype = int)
#create array of ones
arrone = np.ones([3,2], dtype = int)

Try these commands on your editor and let us know what you get.

Numpy is the basic package that is required for other scientific packages like scikit-learn, OpenCV, scipy etc…

2. Matplotlib

It is one of the most common libraries used for the data presentation stage of data science. A comprehensive library that can create static, dynamic, interactive and animated visualizations, matplotlib is very common for creating different types of graphs that represent data using pyplot.

To use matplotlib, install it using the command –

pip install matplotlib

You can then use any functions inside the library to create various graphs. For example,

from matplotlib import pyplot as plt
plt.plot([1,2,3,4], [2, 4.5, 0, 8])


3. Pandas

Pandas, meaning Panel (Pan) Data (da), is a rich library that helps in ETL (Extract-Transform-Load) phase of data science. It is also used in many machine learning and deep learning use cases. The usage of Python for data science drastically increased after the introduction of Pandas library. It contains single and multi-dimensional data structures that are used for data manipulation. For example, you can create a dataframe or series from a list, dictionary or any other collection.

Install pandas library using pip install pandas command.

Try this simple code that converts data into a tabular format, to get started with pandas:

import pandas as pd
pd.DataFrame({"EmployeeId": [12, 13, 14, 15],
"Skill": ['Data science', 'machine learning', 'java', 'python']})


Note that in Python, a sequence is created by default and starts with zero. As you see, a neat structure is created. With this kind of presentation, you can easily perform data manipulation operations like merging datasets, column insertion, deletion & renaming, handling missing values and duplicate data, data sorting, row insertion, renaming and deletion, data exploration, and visualization.

4. Scikit-learn

This free Python library is the magic potion for machine learning algorithms. It contains brilliant tools for data mining and analysis. Algorithms like classification, logistic regression, times series algorithm can be implemented by using sci-kit learn library. Check out the most common examples on the official scikit-learn website.

5. Scipy

Scipy is an important library that is used for scientific computing, mathematics, engineering, etc… There are many useful sub-packages like linalg(linear algebra), interpolate, integrate, stats(statistics), signal, ndimage(image manipulation) etc… Just like Scikit-learn, it is used for the data modelling stage in data science. You can think of Scipy as an advanced version of Numpy, as you can find most data science features in Scipy.

Install Scipy package using pip install scipy command.

Try this simple example of using scipy for linear algebra functions –

from scipy import linalg
#create a square matrix and then calculate the inverse of it
import numpy as np
#our square matrix
twod_matrix = np.array([[2,3], [1,8]])
#now inverse

image rotation using scipy

Try this interesting example of image rotation using scipy –

from scipy import misc, ndimage
from matplotlib import pyplot as plt
import numpy as np
#get face image of panda from misc package
panda = misc.face()
panda_rotate = ndimage.rotate(panda, 180)
plt.imshow( panda_rotate)

fourier transform

Scipy can be used for fourier transform, eigenvalue, eigenvector, integration etc…

6. Tensorflow

It is used in the data evaluation phase of data science. It is an open source end to end library to manage all the aspects of machine learning. The main variant of tensorflow is tf.keras which consists of machine learning algorithms to make predictions. To learn tensorflow, you should know the basics of numpy and pandas. Check out this simple linear regression model done using TensorFlow. The code is written in colab (which is similar to jupyter).

7. Statsmodels

Used for data modelling, testing and analysis, statsmodels is a library that has an extensive list of descriptive statistics, plotting functions, statistical tests, and result statistics for different data types. You can think of this library as a complement to the scipy library. Check the features (models included) of Statsmodels here.

8. Seaborn

Seaborn is an amazing data exploration library based on matplotlib. You can create some beautiful visuals using this library. Install seaborn using the command pip install seaborn in the command prompt (terminal).

Import this library using the command import seaborn as sb

Try this simple time series line plot using sample data that comes with the library.

There are many more features in Python, but the above is good to start with data science. We will now look at specific features that are widely used for data science. As an example, let us try to fetch the data from a file, display it nicely, get some insights, and look at them visually.

Features of Python for data science

If you are overwhelmed with the information presented in various blogs and are confused about how to start using Python for data science, you have come to the right place. In this section, we will go step by step to tackle a simple problem. We won’t use a real dataset, rather create a dataset with few rows – the intention is to understand how we can perform data science. Although we will not get a very accurate result (due to lack of data), by the end of this, you will be able to understand how Python libraries work and how each of them can be used at various stages of data science.

Problem statement – find the best candidates in a firm based on their skills and performance to form a team for the upcoming gaming challenge.

Loading Data with Python

We have created a dataset with few values. First, we need to load the data. This can be easily done using the read_csv() method of the pandas library. If your data is in an excel, you can use read_excel() method.

But, before we do that, we need to import the libraries. We will need pandas (which uses numpy), matplotlib, and seaborn.

import pandas as pd #for data loading, cleaning
import seaborn as sns #data visualization
import matplotlib as mpl #plotting
import matplotlib.pyplot as plt #plotting

#get data from the source
data = pd.read_csv(r'C:\Users\Ramya\student_details.csv')

For the data file, click here. (note that this is not real data, it is just a simulation to get a grasp of important concepts; you can play around with this data as much as you want to.)

Cleaning and sorting the data

Before we do that, we need to make sure that we are able to see the data properly, to know what details we need. Let us use pandas dataframe to present the data in a tabular form.

#create a dataframe (tabular structure)
gameteam = pd.DataFrame(data, columns = ['name', 'age', 'primary_skill', 'secondary_skill', 'income', 'performance', 'availability', 'experience'])
#list the first 10 rows

Example 1

#let us now clean the data, by removing empty, null, and duplicate values
gameteam = gameteam.dropna()

In our data, there are no null or duplicate values as of now, but we will be using this feature again later.

Data Visualization with Python

So, as the first thing, let us view the experience of only those who have experience of more than 3.5 years. You will be amazed at how simple this is.

#for visualization we need to use seaborn and matplotlib
sns.lineplot(x='name', y=(gameteam.experience[gameteam.experience>3.5]), data=gameteam)

You will get a line plot like this –

Data Visualization with Python Graphics

Next, let us create a boxplot to find the five-number summary of the dataset. We will do this for the income and experience values from our data.

sns.boxplot(x = 'experience', y = 'income', data=gameteam)

Graph 2

Let us quickly analyse this boxplot. For example, if you see the x-axis value 4.5 (experience), there are two values in the data set.


Box plot depicts the following values –

Maximum value is represented by the upper whisker = 90000
Minimum value is represented by the bottom whisker = 50000
Median represents the middle value i.e. (90000+50000)/2 = 70000
First quartile (median of the lower half of the dataset) = 60000
Third quartile (median of the upper half of the dataset) = 80000

Read more about box plots.

How about dividing the population into intervals based on experience? For example, separate those who have experience less than some number of years from those who have more.

We can do this using a histogram. By choosing the bin or bucket value, we can decide how to split the data. The value to split is decided based on the max and min value in the distribution.

#let us see the distribution of experience using histogram
#you can change the bin value - as you increase, you will see how the values spread

how the data is spread

Change the bin value to 3 to see how the data is spread into least, average and most experienced candidates.

Next, let’s find out who is available to work for our gaming challenge and has the right skills. For this, we need to filter the data based on some conditions. Suppose, the primary skill required for gaming is C++. We can get names of those who have primary skill as C++ and are available for the challenge.

finalteam = gameteam[['name', 'experience']] =[(gameteam.primary_skill == "C++") & (gameteam.availability == "Yes")]
#remove null entries
finalteam = finalteam.dropna()

Tables 2

We get a list of those who are qualified. We can plot a line plot to understand how varied their knowledge would be based on the experience they have. Let us sort based on the experience from lowest to highest value. This can be done using seaborn’s pointplot().

seaborn’s pointplot()

#sort and then reset index
finalteam = finalteam.sort_values('experience', ascending = True).reset_index(drop=True)
lplt = sns.pointplot(, y=finalteam.experience)

Graph 3

So, that’s about it. We did some quick analysis and visualization of data using Python.

However, learning Python for data science is incomplete without building a predictive model. We need to know which of these candidates are most likely to be selected for the game challenge, which is based on their performance. Since, we have only a few rows, we found out the best performers just by looking at the graph. However, we can get this same information using machine learning algorithms too, which would be much easier, when we have a huge dataset.

Let us define the criteria again –

  • Candidates must have C++ as their primary skill
  • They should be highly performant
  • They should have experience of at least 3 years (optional)
  • They should be available for the challenge

The simplest model we can use for this is the decision tree model, where through a set of questions, we will be able to get the answer to our criterion. Check out our article on the decision tree for machine learning that implements a decision tree in Python.

Which other models do you think can be used for this? Let us know in the comments.

Let us make a decision tree classifier as an example.

Note that as this is an introductory (basic) article, we will not go into correcting over-fitting and other issues with accuracy. In this article, we focus on learning how a model works.

Before passing data to our model, we need to convert string values like primary_skill, availability to numeric values, as sklearn won’t accept any other values.

from sklearn.tree import DecisionTreeClassifier, export_graphviz
from sklearn import metrics
from sklearn.preprocessing import LabelEncoder
#all the values should be numeric, so we need to assign codes to fields like primary_skill
#and availability
def encodeMyFeatures(df):
columnsToEncode = list(df.select_dtypes(include=['category','object']))
le = LabelEncoder()
for feature in columnsToEncode:
df[feature] = le.fit_transform(df[feature])
print('Error encoding '+feature)
return df
gameteam = encodeMyFeatures(gameteam)

values of performance

Now, let’s use the model to predict the values of performance.

model = DecisionTreeClassifier()
predictor_var = ['experience','primary_skill','availability']
#Fit the model:[predictor_var],gameteam['performance'])
#Make predictions on training set:
predictions = model.predict(gameteam[predictor_var])
#Print accuracy
accuracy = metrics.accuracy_score(predictions,gameteam['performance'])
print ("Accuracy : %s" % "{0:.3%}".format(accuracy))

Note that if the data set is small, we would get good accuracy. This is a very simple model built with a lot of things left out – outliers, over-fitting, underfitting etc….

However, through this, you should be able to use Python for data science end-to-end.


Data science is vast and so is Python. However, with just some basics of both, you can focus on learning Python for data science to start working on technical aspects of data science. Python will help you grow as a programmer as well as a data scientist. In this article, we have used packages that cover end to end stages of data science from preparation to visualisation. Check out our complete Python tutorial, where you can start from scratch and get in-depth knowledge of Python for data science and otherwise.

Leave a Reply

Your email address will not be published. Required fields are marked *