Description
Welcome to this course on Python for Data Science. This is a 4 week course we are
going to teach you some very basic programming aspects in python. And since this is a
course that is geared towards data science towards another course based on what has
been taught in the course, we will also show you two different case studies one is what
we call as a function approximation case study another one a classification case study.
And then tell you how to solve those case studies using the programming platform that
you have learned. So, in this first introductory lecture I am just going to talk about why
are we looking at python for data science.
(Refer Slide Time: 01:10)
So, to look at that first we are going to look at what data science is. This is something
that you would have seen in other videos of courses in the NPTEL in other places. Data
science is basically the science of analyzing raw data and deriving insights from this
data. And you could use multiple techniques to derive insights, you could use simple
statistical techniques to derive insights, you could use more complicated and more
sophisticated machine learning techniques to derive insights and so on.
Nonetheless the key focus of data science is in actually deriving these insights using
whatever techniques that you want to use. Now there is a lot of excitement about data
science and this excitement comes because its been shown that you can get very valuable
insights, from large data and you can get insights about how different variables change
together, how one variable affects another variable and so on with large data which is not
very easy to simply see by very simple computation.
So, you need to invest some time and energy, into understanding how you could look at
this data and derive these insights from data. And from utilitarian viewpoint, if you look
at data science in industries if you do proper data science, it allows these industries to
make better decisions. These decisions could be in multiple fields for example,
companies could make better purchasing decisions, better hiring decisions, better
decisions in terms of how to operate their processes and so on.
So, when we talk about decisions, the decisions could be across multiple verticals in an
industry. And data science is not only useful from an industrial perspective it is also
useful in actual science as themselves. So, where you look at lots of data to model your
system or test your hypotheses or theories about systems and so on. So, when we talk
about data science, we start by assuming that we have a large amount of data for the
problem of interest. And we are going to basically look at this data we are going to
inspect the data, we are going to clean and curate the data then we will do some
transformation of the data modeling and so on before we can derive insights that are
valuable to the organization or to test a theory and so on.
(Refer Slide Time: 03:47)
Now, coming to a more practical viewpoint of what we do once we have data. I have
these four bullet points; which roughly tell you supposing you were solving a data
science problem what are the steps you will do? So, you will start with just having data
someone gives you data; and you are trying to derive insights from this data. So, the very
first step is really to bring this data into your system. So, you have to read the data. So,
that the data comes into this programming platform so that you can use this data. Now
data could be in multiple formats so you could have data in a simple excel sheet or some
other format.
So, we will teach you how to pull data in to your programming platform from multiple
data formats. So, that is a first step really if you think about how you are going to solve a
problem these steps would be first to simply read the data. And then once you read the
data many times you have to do some processing with this data you could have data that
that is not correct. For example, we all know that if you have your mobile numbers, there
are 10 numbers in a mobile number and if there is a column of mobile numbers and then
say there is a one row where there are just five numbers then you know there is
something wrong ok. So, this is a very simple check I am talking about in real data
processing this gets much more complicated.
So, once you bring the data in when you try to process this data you are going to get
errors such as this. So, how do you remove such errors how do you clean the data? Is one
activity that that usually precedes doing you more useful stuff with the data. This is not
the only issue that we look at there could be data that is missing.
So, for example, there is a variable for which you get a value in multiple situations, but
in some situations the value is missing. So, what do you do with this data do you throw
the record away? Or you do something to fill the data and so on. So, these are all data
processing cleaning steps. So, in this course we will tell you the tools that are available
in python so that you can do this data processing cleaning and so on.
Now what you have done at this point is you have been able to get the data into the
system, you have been able to process and clean the data and get to a certain data file or
data structure that is reasonably complete so that you think you can work with this data
set at which point what you will do is you will try to summarize this data. And usually
summarization of this data a very simple technique would be very very simple statistical
measures that you will compute; you could for example, computer median, mode, mean
of a particular column.
So, those are simple ideas or summarizing the data you could compute variance and so
on. So, we are going to teach you how to use this notions of statistical quantities that you
can use to summarize the data. Once you summarize the data then another activity which
is usually taken up is what is called visualization right. So, visualization means you look
at this data and more pictorially to get insights about the data before you bring in heavy
duty algorithms to bear on this data. And this is a creative aspect of data science, the
same data could be visualized by multiple people in multiple ways. And some
visualizations are not only I caching, but are also much more informative than other
types of visualization.
So, this notion of plotting this data so that some of the attributes are aspects of the data
are made apparent is this notion of visualization. And there are tools in python that will
teach you in terms of how you visualize this data. So, at this point you have taken the
data, you have cleaned the data, got a set of data points or data structure that you can
work with you have done some basic summary of this data that gives you some insights.
You also looked at it more visually and you have got some more insights, but when you
have large amount of data big data the last step is really deriving those insights which are
not readily apparent either through visualization or through simple summary of data.
So, how do we then go and look at more sophisticated analytics or analysis of data so,
that these insights come out. And that is where machine learning comes and as a part of
this course when you see the progress of this course you will notice that you will go
through all of this, so that you are ready to look at data science problems in a structured
format and then use python as a tool to solve some of these problems.
(Refer Slide Time: 08:57)
Now, why python for doing all of this? The number one reason is that there are these
python libraries, which already are geared towards doing many of the things that we
talked about so that it becomes easy for one to program and very quickly you can get
some interesting outcomes out of what we are trying to do.
So, there are as we talked about in the previous slide, you need to do data manipulation
and pre processing. There are lots of functions libraries in python where you can do data
wrangling manipulation and so on. From a data summary viewpoint there are many of
these statistical calculations such you want to do are already pre programmed and you
have to simply invoke them with your data to be able to show data summary. The next
step we talked about visualization there are libraries in python, which can be used to do
the visualization.
And finally, for the more sophisticated analysis that we talked about all kinds of machine
learning algorithms are already pre coded available as libraries in python. So, again once
you understand some bit about these functions and once you get comfortable working in
python, then applying certain machine learning algorithms for these problems become
trivial. So, you simply call these libraries and then run these algorithms.
(Refer Slide Time: 10:29)
At a higher level so, in the previous slide we talked about flow process for how I get the
data in clean it. And all the way up to insights and then parallelly we said why python
makes it easy for us to do all of this. If you go back if you go forward a little more and
then, ask in terms of the other advantages of python which are little more than just very
simple data science activities. Python provides you several libraries and its being
continuously improved so, anytime there is a new algorithm those are coming into the set
of libraries. So, in that sense its very varied and there is also a good user community.
So, if there are some issues with new libraries and so on and those are fixed so that you
get robust library to work with and we talk about data and data can be of different scale.
So, the examples that you will see in this course are data of reasonably small size, but in
real life problems you are going to look at data which is much larger which we call as big
data. So, python has an ability to integrate with big data frameworks like hadoop spark
and so on.
And python also allows you to do more sophisticated programming object oriented
programming and functional programming. Python with all of this sophisticated tools
and abilities is still reasonably a simple language to learn its reasonably fast to prototype.
And it also gives you the ability to work with data which is in your local machine or in a
cloud and so on. So, these are all things that one looks for when one looks at a
programming platform which is capable of solving problems in real life right.
So, these are real problems that you can solve, these are not only toy examples, but real
applications that you can build data science applications that you can build with python.
(Refer Slide Time: 12:49)
And just as another pointer in terms of why we believe that python is something that, a
lot of our students and professionals in India should learn. As you know there are tools
which are paid tools for machine learning with all of these libraries and so on.
And there are also open source tools and in India based on a survey, most people of
course, prefer open source tools for a variety of reasons cause being one because its free
to use. But also if it is just free to use, but it does not have a robust user community then
its not really very useful that is where python really scores in terms of a robust user
community which can help with people working in python. So, it is both open source and
there is a robust user community, both of which are advantageous for python.
(Refer Slide Time: 13:48).
And if you think of other competing languages for machine learning; if you look at this
chart in India about 44 percent of the people who were surveyed said they use python or
they prefer python. And of course, a close second is R. In fact, R was much more
preferred a few years back, but over the last few years in India a python is starting to
become the programming platform of choice. So, in that sense its a good language to
learn because the opportunities for jobs and so on or lot more when when you are
comfortable with python as a language.
So, with this I will stop this brief introduction on why python for data science. I hope I
have given you an idea of the fact that while we are going to teach you python as a
programming language, please keep in mind that each module that we teach in this is
actually geared towards data science. So, as we teach python we will make the
connections to how you will use some of the things that you are seeing in data science;
and all of this we will culminate with these two case studies that will bring all of these
ideas together. In terms of both giving you an idea and an understanding of how the data
science problem will be solved and also how it will be solved in python which is a
program of choice currently in India.
So, I hope this short four week course, helps you quickly get on to this programming
platform. And then learn data science and then, you can enhance your skills with much
more detailed understanding of both the programming language and data science
techniques.
Thank you.
Now, the commonly used data exploration and visualization tools are Tableau, Qlikview
and of course, you always have your MS Excel. So, the next bucket that we are going to
look into is when you have huge chunks of data, now when your collecting data on a real
time basis you are going to be collecting data over every second every minute. Now if
you want to store all these data and preprocesses it the regular desktop or computing
systems that you have might not be useful.
So, that is when you use parallel or distributed computing, where you distribute the work
across different systems popular tools that are being used for big data apache spark and
Apache Hadoop. So, in this course we are going to be mainly focusing on tools that are
required for data preprocessing and analysis and in specific we are going to look into
python.
(Refer Slide Time: 03:08)
So, let us look at the evolution of python. So, python was developed by Guido van
Rossum in the late eighties at the national research institute for mathematics and
computer science and this institute is located at Netherlands.
So, there are different versions of python, the first version that it was released was in
1991; the second version was released in 2000 and the third version was released in 2008
with version 3.7 being the latest. So, let us look at the advantages of using python.
(Refer Slide Time: 03:41)
So, python has features that make it well suited for data science. So, let us look at what
these features are. So, the first and foremost feature of python is that it is an open source
tool and python community provides immense support and development to its users. So,
python was developed under the open source initiative approved license thereby making
it free to use and distribute even if its for commercial purposes.
(Refer Slide Time: 04:05)
The next feature is that the syntax that python use fairly simple to understand and code
and this breaks all kinds of programming barriers if you are going to switch to a newer
programming language. So, the next important advantage of using python is that, the
libraries which are contained in python get installed at the time of installation and these
libraries are designed keeping in mind specific data science task and activities.
Python also integrates well with most of the cloud platform service providers; and this is
a huge advantage if you are looking to use big data. So, if you are going to download
python from the website and install it, you will see that most of the scripting is done in
shell. So, there are applications that provide better graphical user interfaced for the end
users and these are taken care by the integrated development environment.
(Refer Slide Time: 04:57)
So, now, let us see what an integrated development environment is, an IDE as how its
abbreviated is a software application and it consists of tools which are required for
development. All these tools are consolidated and brought together under one roof inside
the application. IDEs are also designed to simplify the software development this is very
useful because as an end user, if you are not a developer you might want all the tools
available at a single click. Using an IDE will be very beneficial in that case also the
features provided by IDEs include tools for managing compiling deploying and
debugging a software. So, these also form the code features of any IDEs.
(Refer Slide Time: 05:44)
So, now let us look at what are the features of an IDE in depth. So, any IDE should
consist of three important features; the first is the source code or text editor, the second is
a compiler and the third is a debugger. Now all these three features form the crux of any
software development.
The IDEs can also have additional features like syntax and error highlighting code
completion and version control.
Checkout: https://coursetime.net/machine-learning-for-beginners-linear-regression-model-in-r/