21 POWERFUL TIPS,

TRICKS, AND HACKS FOR

DATA SCIENTISTS
 CONTENTS
INTRODUCTION
21 TIPS, TRICKS, AND HACKS FOR DATA SCIENTISTS
A. General Tips and Tricks
B. Python Tips, Tricks, and Hacks
C. Data Extraction Tips, Tricks, and Hacks
D. Data Pre-Processing Tips, Tricks, and Hacks
E. Model Building Tips, Tricks, and Hacks
DEBUNKING COMMON MISCONCEPTIONS
GROWING YOUR SKILLS
ABOUT DASCA
02
INTRODUCTION

Data is the oil of the 21st century, and people So how do these data wizards mine
who can analyze it are the high-performing the terabytes of information out
combustion engines who extract it, refine it, there?
and use it effectively to make it valuable.
This guide elucidates some time-
Jobs in data science are some of the best- saving hacks, tips, and tricks that
suited vocations to thrive in this era. The data scientists can use to make their
most appealing aspect of a career in data lives easier. It will help you make
science is it opens the world to you. Data is smarter business choices and
the new age gatekeeper that can get you into become a better and more efficient
about any arena. data scientist.

From Amazon to Facebook and down to the Data Science is a growing discipline.
thousands of disruptors across industries, There is still a lot to learn for both
data science has become intrinsic to young and seasoned professionals.
business success. Companies are culling Reflecting on the decades of
through their troves of raw information to experience of Senior Data Scientists
gain a competitive advantage. All are seeking and industry best practices, this
a firm footing in data and analytics. guide brings to you:

03
Being a data scientist is not only about data crunching.
It’s about understanding the business challenge, creating
some valuable actionable insights to the data, and
communicating their findings to the business.

JEAN-PAUL ISSON
VP-Chief Data Science & Artificial
Intelligence Officer, SITA
A DATA SCIENTIST

They are the unicorns who bring a creative combination of an engineer, “You can’t simply go to a business stakeholder and say that I ran a
visualization expert, an analyst, a researcher, and a navigator between logistic regression on this data to classify…,” he says. On a related note,
business and data science teams. Heather Nolis, a machine learning engineer at T-Mobile says that for a
better working relationship with engineers, data scientists must
Data Scientists come from a diverse range of backgrounds – from appreciate and try to speak the language of engineers.
engineering and mathematics to business and even social sciences. With a
confluence of quantitative and analytical skills, they play a central role in They are, indeed, the bridge between technical and business; not only
providing strategic guidance with business intelligence. It’s one of those by function but by training.
job roles that may sound super-technical, mysterious, and even hard to
get, but with the right push in the right direction, one can become a good Knowledge and awareness are the greatest strengths of a good data
data scientist. scientist. No matter how much you can code or how good you are at
statistics – your business awareness and familiarity with uncertainty
Javis Miller, a data scientist at Spotify shares that data scientists worth are the most valuable skills you can offer as a data scientist.
their salt must be able to explain their contribution to business without
using jargon.

05
 DATA SCIENTISTS EMPLOYERS SEEK

Business acumen Collaboration Communication

In-demand Soft-Skills Attention to detail Critical thinking

Storytelling

Leadership Project
Problem Solving Management

Statistics Data Modelling Data Visualization

In-demand Hard-Skills Machine Learning

Mathematics R

Natural Language SQL

Python Processing
06
21 TIPS, TRICKS, AND HACKS FOR DATA SCIENTISTS

This section is divided into five parts, representing the major phases of the data science process. Part one covers a few general tips important for data
scientists, followed by Python, data extraction, and data preparation, and finally predictive modeling.

Data science has found its way into all industries and disciplines. Find some basic, yet critical tips and tricks to
General understand the strategies and practices followed by experts in data science. These tips will help you focus on what's
necessary and discard the unnecessary.

Python is the go-to language in data science and machine learning. Learn some 'pythonic' hacks to code better and
Python easier. Given the universality of the language among data scientists, all the examples in this guide follow Python
coding representation.

Extracting data is the first step to bring your hypothesis to reality. Your ability to tell from where to aggregate the data
Data and the skills to review it to eliminate the errors can make or break your model. Learn some tips and tricks to extract
Extraction and clean the data to make it analysis-ready.

Data preprocessing is an important step before model building. Normalizing the data, standardization, categorizing
Data Pre- variables are just a few tasks performed at this stage. Learn a few hacks on the functions to use for normalization and
processing
testing if the data you have is enough.

Building models is an important stage for data scientists. It guides the process of actually architecting the data for the
Model business context. In these model building hacks, tips, and tricks explore the libraries you can use for robust model
Building testing and data visualization.
07
 General

1. ASK THE RIGHT QUESTIONS TO SET GOALS

A data set will tell you no more than what you ask of it. Nurture and improve your knack for asking the right questions that lead to informative answers.

One way is to begin by investigating what do the customers expect. It may be related to a much-needed product improvement or a problem. Often customer
expectations can lead to the questions and the answers that you must seek from the data.

A word of caution as you ask your questions: always acknowledge your assumptions and make sure they are correct. Your results will be as good as your
assumptions. Recognizing your assumptions can be a challenging feat as often they are innate to our knowledge. Accomplish this by breaking the reasoning
between your observation and conclusion into logical steps. For instance:
Relevant
Observation: A has gone down recently. Data
Assumption: Because A always course-correct itself Good
to a certain mean value, X, Question
USEFUL
Conclusion: A will soon go up toward X
ANSWER

Insightful
Analysis 08
 General

2. SET TIME ASIDE FOR EXPLORATION OF NEW LIBRARIES AND FEATURES

Life can be busy as you move into a job. As a data scientist, make sure you intently take out time for leisure discoveries of new tools and updates that may not be
directly related to your regular job. Going out of your comfort zone will unleash your imagination, keep your knowledge up-to-date, and can help you prepare
yourself for uncertainties.

You could explore new libraries and find some incredible tools that save you time. Pandas Profiling, for instance, can be a great package when working with smaller
datasets. It conducts exploratory data analysis and converts it into a report. All in one line of code! Gradio is another package that helps you develop and deploy a
web app for ML model in just three lines of code. These are new developments. Unless you take time out to discover the new releases and integrate yourself into
data science community, it will be difficult to find these gems.

Top 10 Data Science Tools You Should Know:

1 2 3 4 5 6 7 8 9 10

NLTK D3js H2O BigML SparkML WEKA Hive MATLAB Tableau Mongo
DB

09
 General

3. FOLLOW THE BEST CODING PRACTICES

It’s always advisable to follow the best practices for code documentation and preservation for your and others’ sake lest someone else has to work on your
project and code in the future. Make sure everyone in your team abide by these rules. Here are a few best practices you must follow.

Use proper documentation to write Make sure your variables, files, Consider naming a scheme and
Add comments in between your
a short description of what the code folders, and documents have sticking to it throughout your code.
code to make it legible for fellow
and the project are about and how meaningful names. (It’s one of the
programmers.
to use it. best ways to stay organized.)

Try to import all libraries at the If your code is becoming too heavy, Use versioning and code repositories.
Maintain proper spacing between
beginning of your notebook. (You you can hide a part of it that’s not Modern repos are based on versioning
your code lines.
can add a comment on why you are important right now. It will make systems, which track your code for the
loading each.) your notebook cleaner. changes you make.

10
 General

4. USE GIT AND GITHUB

From learning version control to building your data science portfolio, Git and GitHub are the important resources you must develop a good working knowledge of.
Github.com can provide you free web hosting of code repos in a version control system. It also facilitates collaboration among programmers. You can decide to
keep it public or private. You can find online tutorials on how to get started. Many seasoned data scientists also use GitHub repositories to demonstrate their
capabilities to do a job by showcasing their contribution through GitHub.

TOP 15 MOST VALUABLE GITHUB REPOS IN 2021

kamranahmedse / jlevy/the-art-of- thedaviddias/Front airbnb/ FrontendMasters/l sdmg15/Best-websites-a- getify / You-Dont-

developer-roadmap command-line -End-Checklist javascript earning-roadmap programmer-should-visit Know-JS

bradtraversy/desig anuraghazra / mbeaudru/modern ryanmcdermott/cle trekhleb/javascript microsoft/Web- sindresorhus / EbookFoundation /

free-programming- 11
n-resources-for- github-readme- -js-cheatsheet an-code-javascript -algorithms Dev-For-Beginners awesome
developers stats books
 General

5. KNOW THE MOST IMPORTANT LANGUAGES – R, PYTHON, SQL, AND

MATLAB

R, Python, SQL, and MATLAB are a must-know for every programmer: data analyst or data scientist. A programming language enables you to design programs specific
to your needs. You can then reuse these programs for projects whenever you require taking the same action.

R and Python are two of the most important programming languages used for all purposes. To begin with, you can choose either of them to develop your programming
skills. Their biggest advantages are:
They can manipulate data and work with multiple data types and software, and
They can give end-to-end solutions for business problems.

One of the limitations of R and Python, though, is they cannot work with specific domains such as relational database management systems (RDBMS). SQL is used for
that purpose. It can be used for querying from RDBMS and also in preparation for BI analysis. This makes SQL a must-know language for data scientists.

Another language that is essential to data science is MATLAB. It is used for working with matrices and tables. However, it is a paid service, which is its biggest limitation.
Either way, these four languages cover most of the languages and tools used when working with data.

That said, when dealing with Big Data, apart from R and Python, faster languages like Java and Scala are often used. The latter two are very useful when combining data
from a variety of sources.

Some important software: Software is usually created by a combination of these languages and provides easy to use functionalities (check Pro Tip on the sidebar).

MS Excel and SPSS are some examples of software used in analytics. For working with Big Data, applications such as Apache Hadoop and MongoDB are used. Power BI,
Tableau, SAS, and Qlik are some of the top-notch BI visualization software.
12
 Python

6. USE IPYTHON OR JUPYTER NOTEBOOK FOR IDE

Once you install Python, you will need to pick a virtual or an integrated development environment (IDE). A virtual environment is one of the most important
tools when working with Python. It keeps the dependencies of different projects separate by creating isolated virtual environments.

There are three options you can choose from:

IDLE (a default Jypter Notebook

Terminal
environment) (earlier called
iPython)

We highly recommend that every data scientist use Jupyter Notebooks. From codes to presentations and generating full-fledged reports, Jupyter Notebooks
provide a range of amazing features. Say, you write your code, you can run it in blocks than going line by line. It can also be used to perform data visualization. To
create a presentation, you can create slides with super flexibility. To convert your notebook into slides, all you need is to choose Slideshow from Cell Toolbar. It
has a powerful, versatile, and shareable interface. Jupyter IDE truly helps data scientists do it all.

13
 Python

7. KNOW THE 12 PYTHON LIBRARIES YOU WILL NEED

Python offers a gamut of useful libraries. Here are the libraries that data scientists use the most for analysis and computations. A few of them are also used in
the examples used in the rest of this guide.

NumPy SciPy Pandas Matplotlib Scikit- Statsmodels

Learn
Working with basic linear Working with high-level Data preparation and Plotting graphs in varied Machine Learning; Statistical Statistical modeling;
algebra functions, engineering modules such munging, especially forms like histograms or Modeling Exploring data; Performing
advanced random as optimization matrices, structured data operations heat plots. You can use its statistical tests; Working
numbers; and integrate and discrete Fourier Pylab function in Jupyter with different data types
low-level languages like C transform. Also used as an Notebooks and convert your
and C++; n-dimensional add-on library to NumPy virtual environment to like
array (best feature) that of MATLAB.

Bokesh Blaze Requests Scarpy SymPy Seaborn

Creating interactive plots Making effective interactive Access the web. It’s more Web crawling; Discerning Symbolic computations Data Visualization and
and dashboards for web dashboards and visuals with convenient and easier to data set patterns and requiring a lot of making informative and
browsers; Working with large datasets code compared to other digging throughout the capabilities – from beautiful statistical graphics
huge streaming or huge similar libraries. webpages to find calculus to basic
datasets information arithmetic

14
 Python

8. CREATE A ONE-LINE FUNCTION WITHOUT DEF KEYWORD (LAMBDA)

There may be times when you need to use a simple function for one time in the larger Python program. In such cases, going through the process of defining a
function using def keyword, determining its name, and so on may look like a hassle.

If you are sure that you won’t need to use this function after a certain program, you can use a shorter syntax that will allow you to focus on the function
instead of other formalities associated with def keyword function. You can do it by using the anonymous function, also referred to as Lambda. Here’s how the
both differs.

A DEF KEYWORD FUNCTION: A LAMBDA FUNCTION:

def raise_to_the_power_of_3(x); raise_to_the_power_of_3 = lambda x: x ** 3

return x ** 3

15
 Python

9. FORMAT YOUR CODE USING BLACK

Black is the uncompromising code formatter! If you are developing a lot of code, it can make your life so much easier by cleaning your code and making it easy
to read for you and others. It also makes code reviewing faster.

Black is an automatic formatter for Python. All you have to do is write the code and then let Black format it. This is a great way to focus on your code instead
of your structure. It’s like waking up to a house that magically cleans itself. To invoke the power of Black code formatter:

Save your file with .py extension

On your terminal type “black [filename.py]”

That’s it! You have an automatically formatted file

16
 Data
Extraction

10. KNOW YOUR DATA SOURCES

Albeit the playground of data engineers and analysts, data scientists are always concerned with scouting for data. Understanding the type of data available
and how to store them is fundamental to data science. As a data scientist, your concerns would be: Where is the data now? In what shape is it? And How can I
access it?

Data can be broadly divided into three forms. Structured, which includes transactional data as on relational database management systems. Its ready-made
structure makes it easiest to work with. Semi-structured including text files, XML files, HTML files, among others that requires conversions into forms to
makes it easier for analysis. Finally, Unstructured data, which includes audio, video feeds, web pages, and sensor data. It is the hardest to convert.

Additionally, there are three ways in which data scientists can access the data: File, Database, or API. In all three cases data may be stored in any format: flat
files, XML, JSON, HTML, etc. Each type has its properties and qualities. Google search, Company websites, Kaggle, and such sources can help you scout for the
data you need. In web searches, keep it in mind to use keywords such as ‘data’ and ‘API’. It can make a substantial difference.

NON-RELATIONAL
DELIMITED RDBMS FLAT FILE DATABASES (WITH JSON XML
FLAT FILE TEXT SEARCH)

17
 Data
Extraction

11. DON’T UNDERESTIMATE SQL

Often data science professionals underrate the importance of learning SQL. Data Scientists often work with databases, and fetching the data can become
tiresome without a knowledge of SQL. It is the most prominent programming language to manipulate data and extract it. What more? It can be of value to a
variety of data professionals – from data engineers and analysts to business analysts and data scientists.

To give a crude example, say you want to analyze millions of customer orders in your company and forecast how the orders per day will change in the time to
come. To do that, first, you will need to write a SQL query to fetch the orders placed each day. Then will come the use of R or Python to run a statistical
forecast. It can be difficult to reach far in data analysis without knowing SQL. Along with RDBMS, SQL also works for new-age database systems such as
MySQL, Oracle, Microsoft Access, IBM, among others.

SQL can be used to:

Set Delete or Retrieve

Make new data from
permissions insert or
tables and and execute Build data
on views and update
new queries pipelines.
tables in a records in a
databases. against
database. database.
databases.

19
 Data
Extraction

12. EXTRACT BY DATATYPE INTO DIFFERENT DATA FRAMES

Data Frames are a two-dimensional tabular representation of data, just like a spreadsheet with columns and rows. ‘Column names’ refer to the columns, and
rows can be accessed using numbers 0 to n. Your data will be first read into a data frame, and then will become ready for other operations. Data frames are
fundamental data models for pandas.

Extracting categorical data (object like name or ID) from numerical data (float and integers like income) into different data frames can be a useful feature at
the time of data visualization and analysis. Here is a simple representation of how you can segregate your data by its type and represent it into different data
frames in Python using Pandas.

Output data frame Output data frame

Import pandas Import the dataset Check data type with only categorical containing only
variable numerical variables
data =
import pandas data.dtypes
pd.read_csv(‘x’) categorical_var = numerical_var =
as pd
data.shape data.select_dtypes(“obj data.select_dtypes(“nu
ect”).head() display mber”) display
(categorical_var.head()) (numerical_var.head())

20
 Data
Extraction

13. REMOVE EMOJIS FROM TEXT

Once you extract your data, the first step is to clean the data to make it more understandable. The first step toward that is to remove unwanted characters such as
emojis, links, unnecessary spaces, punctuations, among others. Given below is an easy way to remove emojis from your text data.

text= "Hi ! Is the group meeting tomorrow? I need to plan something.

So, please let me know. "

processed_text=text.encode(‘ascii’, ‘ignore’).decode(‘ascii’)

#Display text with emoji

print (“Raw message:”, text)

#Display text without emoji

print (“Processed message:”, processed_text)

21
 Data Pre-
Processing

14. TEST YOUR DATA TO KNOW IF IT’S ENOUGH

Say you have the data; now how do you know if that’s enough? Must you keep looking for more or should you attack your data and start with the processing
and analysis?

To figure out the answer to this dilemma take a deeper look at your data. Often data sets aren’t what they seem or what you imagine them to be.

For instance, say you have a goal of recommending perfumes to users of a perfume website based on the ratings the other users have provided for the
perfumes. You begin with the question/hypothesis: do users like perfumes of certain scent types significantly more than others?

For this example, say you get access to a data set from a popular perfume-rating website wherein thousands of users have given ratings of one to five stars.
Now, as you get to test your hypothesis with this data, you realize that the CSV file contains only three columns: USER_NAME, PERFUME_NAME, and RATING.
(Drat! No scent types.) The dataset that may have seemed very relevant would now seem less so. Thus, to find the answer, you must find either a data set
matching perfume names to scent types or try to infer it from the data you have by searching manually from each perfume name.

To find out if your data is enough, before spending much time manipulating your data or diving into analysis, perform a spot check. To do this, pick a few data
points and attempt to arrive at the answer you are looking for with a quick analysis on a small scale. It will save you time, energy, and resources.

22
 Data Pre-
Processing

15. USE .FORMAT() TO PRE-PROCESS TEXT DATA

Python string formatting method .format() is a very useful tool to pre-process text data. It can handle complicated string formatting quite efficiently. It has
a built-in string class, which provides functionality for complex substitutions of variables and formatting of values (integer, floating-point, string,
characters, etc.). The general syntax for using it is {}.format(value). Here’s an example using the format method for single and multiple formatters.

#using format method with multiple parameters

#using format method in simple string with single parameter
my_string = “{} certifications for {} are available across {}
Print(“{}, Nurturing the world’s top-league of data science
countries.”
professionals.”.format(“DASCA”))
Print (my_string.format(“DASCA”, “Data Scientists”, 183))

The above program will give the following output: The above program will give the following output:

DASCA, Nurturing the world’s top-league of data DASCA certifications for Data Scientists are available
science professionals. across 183 countries.

23
 Data Pre-
Processing

16. TRANSFORM DATA INTO A NORMAL DISTRIBUTION

For data to follow a normal distribution pattern is of utmost importance for data scientists, especially if you are trying to do linear and logistic regressions. It is so
because normally distributed data indicates the correctness of the researcher’s assumptions, and makes the model more reliable. Additionally, most statistical
calculations are done with the assumption that the data is normally distributed.

While there are numerous ways to test if your data is normally distributed using statistical software, here are a few transformations you can try on your numeric
features to bring your data closer to normal or gaussian distribution using Python.

Transforming a distribution entails applying functions to the values such that it fixes the skewed data so that the output comes closer to what you expect. One of
the most common goals of transforming the data is to bring it closer to normal distribution. Check out more on transforming data into normal distribution here.

Logarithm np.log(x) and Exponential Inverse 1/x, square root np.sqrt(x), and
np.exp(x) cube root x**(1.0/3.0), and the likes

Polynomial transformation like

x**2, x**3, and so on

24
 Data Pre-
Processing

17. PLOT DECISION TREES

Decision trees provide an intriguing way to extract value from data points. A decision tree is a sequence of yes or no questions, which eventually end in making a decision. It is
a non-parametric machine learning modeling technique often used for classification and regression problems.

Say, a bank needs to decide if it should give someone a loan or not. To arrive at a decision, they will walk through a series of questions such as “What’s the income of the
person?” If the answer is between $100 and $70,000, the bank could choose to continue to the next question, say “How long the person has been working in the current job?”
and so on. Decision trees create a sequential and hierarchical decision process wherein decisional outcome varies with different data inputs.

Using sklearn, you can create a decision tree with just one line of code as shown below. The hyperparameters can be changed as per your needs.

#Sklearn function in Python to plot decision tree

plot_tree()

25
 Model
Building

18. SCIKIT-LEARN IS MOST POPULAR FOR PREDICTIVE MODELING

Scikit-Learn (also known as sklearn) is a strong Python library with immense applicability for data scientists. It is used for machine learning and is built on Scipy, NumPy, and
matplotlib libraries. This gives it functionalities in optimization, clustering, supervised learning, classification, regression, and many other techniques. It is perhaps the most
useful library in Python. You must remember that the Scikit-learn library requires all the inputs in numeric, so you may have to convert your variables into numerical. Here are
a few features Scikit-learn is popular for:

Dataset
Visualization
Unsupervised transformation
learning
Supervised Inspection
algorithms Model Model
learning
selection and Persistence
algorithms
evaluation

From linear Scikit-learn Sklearn offers Scikit defines The inspection Scikit-learn By training a
regression, SVM, supports a gamut many features and an API for module of sklearn libraries can help Scikit-learn
decision trees to of unsupervised methods to select creating helps you clean and model, you can
Bayesian methods, machine learning the metrics and visualizations. understanding preprocess your make the model
the sklearn toolbox algorithms check the accuracy Some of the predictions from a data, reduce it, persist in the
supports all types of ranging from of supervised best features model, and figure expand it, and future without
supervised machine clustering, models. It can be of this API are out what affects even extract having to retrain
learning algorithms. Gaussian mixture used for cross quick plotting them. It can be features from it. You can do it
Its use in supervised model, factor validation for and used to assess texts and with Python
learning is one of the analysis, PCA to evaluating visualization your assumptions images. specific
biggest reasons for neural networks. estimator adjustments. and biases about serialization and
its wide-spread performance or a model and Interoperable
usage. plotting scores to design a better formats.
evaluate models. model.

26
 Model
Building

19. IMPORTING PANDAS IN THE PROGRAM

Pandas has become a very popular library for data analysis. Not to mention, its huge role in ramping up the popularity of Python in data science.

It is used for data manipulation and structured data operations. Pandas has two data structures: Data Frames and Series. Series is a one-dimensional indexed
and labeled array, which is used to access individual elements in your series with labels. Data Frames are columns and rows similar to Excel workbook.

Seasoned data scientists say that pandas data frames work like in-memory and in-Python data stores. So, if you have data that’s big enough for your
computations, but small enough for your computer, then pandas might just be right for you.

Importing Pandas in the program

# printing series values # printing series indexes

# importing pandas in the program # Defining a series object
print("The Series values are:") print("\nThe Index values are:")
import pandas as pd srs = pd.Series([1,2,3,4,5])
print(srs.values) print(srs.index.values)

27
 Model
Building

20. CREATE HEATMAPS ON PANDAS DATA FRAME

Heatmaps are simple. It represents numbers with colors. Say, higher numbers may be represented by warm colors in shades of red and lower numbers may be
represented by cool colors in the shades of blue. Heatmaps have a better impact than conventional visualization techniques as they make spotting patterns easier.
They will help you better understand your range of values in a glimpse.

#import libraries

import pandas as pd

import mnatplotlib.pyplot as plt

import seaborn as sb %matplotlib inline

#define figure size for output plot

fig, ax = plt.subplots(figsize = (12, 7))

Here’s a simple code for creating #define index and columns for data frame
heatmaps on pandas data frame
idx = [‘1’, ‘2’, ‘3’, ‘4’]
using Seaborn
cols = list(‘ABCD’)

#enter values in the index and columns as a panda data frame

df = pd.DataFrame([20, 30, 40, 50], [60, 70, 80, 90], [25, 35, 45, 15],

[5, 15, 10, 5]], columns = cols,index = idx)

#display dataframe as heatmap

sb.heatmap(df, cmap =’RdYlGn’, linewidths = 0.50, annot = True)

28
 Model
Building

21. PLOT ROC CURVES TO ENSURE MODEL INTEGRITY

ROC Curve
ROC curve analysis is critical in both data science and statistics. It shows the performance of a TPR
model by assessing total sensitivity against the fall-out rate. Plotting ROC curve can reveal the
model’s viability. You may be surprised to know that it had immense applicability in WWII wherein it
was used to detect aircraft of the enemy. Since then it has found uses in biology and other
sciences.

Knowing the odds of corrections of predictions is among the main goals of data scientists. Since
predictions are just guessing, a ROC curve analysis can give weight to your predictions. You can
see how accurate are your model’s predictions. In modeling algorithms, the ROC curve is typically
the first test performed. It can detect problems early and let you know if you are on the right path.

It plots the false alarm rate against the hit rate. ROC curve is a plot of the false positive rate (FPR)
against the true positive rate (TPR) for a host of values between 0 and 1. FPR is calculated by
dividing the number of false positives vy the total number of negatives (false positives and true
negatives). Similarly, TPR is calculated by dividing the number of true positives by the total
number of positives. These values are plotted on the curve with FPR on x-axis and TPR on y-axis.
The further the curve from the line, the more predictive is your model.
FPR

29
DEBUNKING COMMON MISCONCEPTIONS

With the overflowing bandwagon of data science, many do not yet fully comprehend the frequently used terms and buzzwords in the field. It has led to the
proliferation of several myths and misconceptions among data science professionals. Here’s to refute and clarify some of the most common misconceptions.

#1 #2 #3 #4

Not all Business It’s not all AI and ML

data is Intelligence about Python are still
Big Data is different and R in data growing as
from Business science disciplines
Analytics

30
1
Calling 3000 lines of data as Big Data is not correct. You must consider several factors
Not all before you dub a data set as Big Data. This is important to understand as the kind of
data you think you are dealing with also has a bearing on the tools you will choose for
data is analysis. Volume is just one of the parameters of defining Big Data. It is also
characterized by the types of data that are involved (variety) and perhaps the variability
Big Data of the sources. Additionally, the velocity with which the information is being received
and being processed. Therefore, never dub a data set as Big Data just by one factor
alone.

2
Business
Intelligence is
A few understand this, but there is a significant difference between business
intelligence (BI) and business analytics (BA). They are not interchangeable terms.

different from Furthermore, not all data analysis can be considered business intelligence. BI deals
with the explanation of past events and future predictions using data-driven

Business approaches and reaching data-based conclusions.

Analytics
31
 Data Science is often associated with strict coding practices and super quantitative skills. It’s

3
widely believed that data science is all about mastering R and Python. Although, it is not
always so.
It’s not all
about Python
Software like Excel, STADA, SPSS, and Tableau are quite frequently used by them.
Sometimes, even more than R and Python. Storytelling is among the many other crucial

and R in data skills of data scientists. All data scientists must have the stomach to process and express
complex mathematical and programming concepts in a story-like presentation to the

science stakeholders. So, if you understand data, know statistical software, and are good at
storytelling, then to begin with you could write code using libraries and you would still be
good to go. Data Science, in essence, is a combination of many skills – from mathematics,
programming, and business. Overemphasizing just one would be a mistake.

4
The talks around Artificial Intelligence and Machine Learning gather much attention.
AI and ML However, you must understand that they are not so old disciplines. The lightning
progress in these domains is a phenomenon of very recent developments, and many
are still developments are yet to happen. Scientists, academics, and researchers still debate at
length about basic functionalities. To give an example, it is widely known that deep
growing as learning can make a model perform exceptionally well. Albeit, it is still not clear how a
machine can obtain those levels of results given existing capabilities. So, as you use ML
disciplines and DL algorithms and models, know that there is much that we don’t know.

32
GROWING YOUR SKILLS

The secret to a great data science career is incessant learning, rich qualifications, and advanced skillsets. If you are looking to take your career up by several
orbits, here are some ways to gain the relevant skills and knowledge to successfully grow in and demonstrate your fitness for data science

Bootcamps
Professional On-the-job Building
and online
certifications training portfolio
courses

33
1
The pre-requisite skills for data scientists fall under three categories – programming and
Bootcamps hacking, mathematics and statistics, and business acumen. You can develop your competencies
by leveraging online resources such as video tutorials and online courses and attending
and online bootcamps. Bootcamps offer a fast-tracked intensive method of training. You could also learn by
reading books authored by seasoned and successful data science professionals.
courses

2 Professional
As you start finding jobs, you’ll realize that demonstrating your skillsets for data science during the
interview is tougher than it seems. Given the scores of competitors eyeing for the same job, you

certifications will need to stand out. Earning a professional certification in data science can demonstrate your
promise and potential for the job, and make you more noticeable to the hiring teams.

Professional certifications are designed to assess aspirants on strict professional standards. Acing
them speak volumes about your fit for the job role. DASCA offers three rungs of prestigious
international credentials in data science for Data Engineers, Data Analysts, and Data Scientists.
Depending on where you are in your professional journey, they can prove to be great career-
shaper qualifications for you. You can contact us to find which qualification will suit you the most.
Explore more here.

34
One thing I would strongly emphasize is that you need to
demonstrate you can do this job. Some ways I have seen
people demonstrate their skills and capabilities is through
open source contributions, speaking at local meetups on
projects they’ve done, and developing a portfolio of
projects. For me, I took a combination of all.

JULIA SILGE
Data scientist and software
engineer, RStudio
 Experience plays a big role in getting a job. The best way to get it is by

3
training yourself in data science while you work. Since work is where we
spend most of our time, you can find the projects at your company that may
On-the-job benefit from a data-driven approach. As you take them up, it will help you
gain on-the-job expertise in data science. Check if your leadership is
training encouraging toward it. You could also assess if your company can sponsor for
you formal training to develop your skills. This route, of course, will mean
more workload than usual, but if you are motivated enough, both you and
your company can benefit from this arrangement.

Networking with fellow data scientists and building your own data science

4
projects can make a world of difference in your professional standing. They
both demonstrate your passion for learning and can give you an edge over
Building others. You can start by making a GitHub profile. Every strong portfolio has
two parts: GitHub repo, which hosts the code for the project, and a blog,
portfolio which explains what you did and why, and gives you an opportunity to show
off your communication skills.

Additionally, if you make your code public, you can interact with other data
science enthusiasts. You can collaborate with them on a few pet projects and
also seek
guidance from your peers about how to grow in the data science field. Kaggle
is another popular data science community platform you can explore.

36
ABOUT DASCA

The Data Science Council of America (DASCA) researches, designs, and builds platform-independent Data Science knowledge frameworks, standards, and
credentials, and certifies individuals entering or working across the spectrum of emerging Data Science professions. The prime goal of DASCA is to develop
high-quality professionals who can squarely address the challenging expectations of Data Science stakeholders internationally.

DASCA offers the world’s most powerful set of credentials along three critical profession-tracks in data science – Big Data Analytics, Big Data Engineering,
and Data Scientists.

DASCA credentials validate the promise and potential of professionals to hit the ground running in the most demanding assignments and roles. DASCA-
certified individuals bring to the table unmatched understanding and capabilities to anticipate and appreciate the need for deploying the latest Data Science
techniques, tools, and concepts to manage and harness Big Data across verticals, environments, and markets.

37
For more information, please visit www.dasca.org

© Data Science Council of America. 2021. All Rights Reserved.