DATA SCIENTIST – HOW TO START IN 8 STEPS

Eager to explore the universe of Data Science and don't know where to start? We will help you. We have prepared a guide that will help you understand what a Data Scientist does and how to start your preparation! Check it out.

McKinsey Global Institute

Big Data Report 2015

"By 2018, there will be a deficit of 140 to 190 thousand professionals with data analysis skills and more than 1.5 million managers and analysts who know how to use Big Data effectively for decision-making."

Let's start by defining what a Data Scientist is:

They are also a sign of modern times. Data scientists were not on the radar a decade ago, but their sudden popularity reflects how companies now think about Big Data. This incredible mass of unstructured information can no longer be ignored and forgotten. It is a virtual gold mine that helps increase revenue – as long as there is someone to dig and unearth business insights that no one had thought to look for. Enter the Data Scientist.

For the general community, a Data Scientist is one of those “Data Wizards” who can acquire masses of data from various sources and then clean, treat, organize, and prepare the data; and then exploit their skills in Mathematics, Statistics, and Machine Learning to discover hidden business insights and generate intelligence.

The data used by a Data Scientist can be both structured (transactional databases from ERP or CRM systems, for example) and unstructured (emails, images, videos, or social media data). The Data Scientist creates algorithms to extract insights from this data. Then, it is up to the Data Scientist to present this data so that decision-makers can use the results of the analysis when defining business strategies or even to create new data-driven products or services.

According to Anjul Bhambhri, former Vice President of Big Data at IBM and current Vice President at Adobe, the Data Scientist is the professional capable of bringing change to an organization through the analysis of diverse data sources. Anjul Bhambhri writes:

There is no specific degree that prepares Data Scientists, as this is a relatively new profession. There has been much debate about this in academia (especially in the US), as the market needs professionals now and the preparation time for a professional with these skills can take some time. Therefore, technical training in certain areas can help prepare these professionals.

With so much information about the Data Scientist profession and its exponential growth in recent years, it is easy to get lost in the face of so many articles and materials with magic formulas on which path to follow. Let me give you a warning: there is no easy path to becoming a Data Scientist! You need to study, learn different techniques, and have interdisciplinary knowledge. For this reason, Data Scientists are well-compensated and hard to find in the market.

Below are the 8 steps we consider fundamental for preparing a Data Scientist:

Step 1: Do a self-assessment

This is the first step and, believe me, it is fundamental. You, as a professional, need to evaluate the current moment of your career and where you intend to be in 5 or 10 years. If you plan to pursue a career in Analytics, whether as a Data Scientist, Data Engineer, or Analyst, you need to understand what your current skills are, where you want to go, evaluate the gaps, and draw up an action plan.

Since there is no specific academic degree to become a Data Scientist, this professional can come from fields such as Statistics or Computer Science, and it is common to find professionals from other fields working as Data Scientists (Marketing, Economics, Social Sciences, etc.). But regardless of the background, some characteristics will be common to all professionals working with Data Science:

Programming – Knowledge of programming is necessary. Programming languages like R, Python, Julia, Scala, and Java are part of the arsenal of tools used in Data Science. Even other data analysis packages, such as SAS, Matlab, Octave, SPSS, and even IBM Watson Analytics, require programming knowledge to extract the best from each tool. It is the programming skill that allows the Data Scientist to put their creativity into practice and extract answers from data to questions that have not yet been asked. If you already have programming knowledge, that will be an advantage. If you don't have programming experience but have a good grasp of the concepts involved in computer programming, that will help a lot. Clearly evaluate your level of programming knowledge.

Logical Thinking – Data Scientists use logical thinking to perform analyses. Programming requires logic. If you already have this skill, it will accelerate your learning in Data Science.

Skill with Numbers – Mathematics is the foundation of Data Science. Computer programming involves skill with numbers. Machine Learning algorithms are based on mathematical concepts. Statistics, a fundamental part of Data Science, requires skill with numbers. Evaluate your characteristics and, in your self-assessment, check if this item will be a problem or not.

Database Knowledge – At various stages of the data analysis process, interactions with databases will be necessary. Relational databases, Data Warehouses, NoSQL databases, Hadoop, SQL language. All of these technologies are directly linked to the Data Scientist's work, and at least understanding them will be a point that can make a difference. Evaluate whether you understand the concept of databases, understand the differences between relational and NoSQL databases, and how to use SQL language for queries.

At this point, you might already be wondering: how do you expect me to learn all of this? Here comes one of the misunderstood concepts about the Data Scientist profession. It is believed that this professional needs to know every tool. This is not true and not even necessary. Choose your tools and specialize in them. For example: knowledge of the R language and Hadoop will allow you to analyze large volumes of data (Big Data). You do not need to know all programming languages, nor do you have to know all databases. The most important thing is logical thinking, which is indeed indispensable (and you might already have this skill). Technology offers tools, and none of them solves 100% of the problems on its own, as they all have their limitations.

By doing this self-assessment, it will be possible to understand your current level of knowledge and start thinking about the action plan!

Step 2: Prepare your computer

Surprised by this step? This is the stage where you prepare your testing environment, and it should not be underestimated. It can be frustrating during your learning process not to have the ideal equipment to install software or run operations that require computational power.

Data science is computationally intensive (this shouldn't be news to you!). Therefore, you need a computer that allows you to process your scripts and learn about data analysis. In addition, you will need to install tools, interpreters, office suites, etc. To work with Data Science, a computer with 8GB of RAM, with an Intel i5/i7 processor or equivalent is our recommendation. Naturally, the higher your computer's capacity, the better! It is also possible to use services like Cloud9 or Amazon AWS and set up a virtual working environment.

Operating System – The decision on which operating system to use is quite personal, and any of the 3 main operating systems (Windows, Mac OS, and Linux) will meet your needs. In any case, you can install virtual machines with another operating system. A large part of the Data Science and Big Data framework was built on Unix platforms. For a Hadoop or Spark server, a Linux server is the best recommendation. For the data presentation part, Microsoft Office and other visualization tools may depend on a Windows system. There is no rule here, but for more advanced users, a Unix system is recommended. For those who feel more comfortable with Windows, there is no problem at all. Use Windows as your operating system and, if necessary, create a virtual machine with Linux if you want to process files with Hadoop and/or Spark or perform other tests. The main Hadoop vendors (Cloudera, Hortonworks, and MapR) provide free virtual machines with Linux and Hadoop, ready to use in a few clicks. Microsoft Azure Machine Learning can be used online, for example. You can also download the SAS University Edition, a virtual machine with Linux and SAS, which in a few seconds allows you to use SAS (one of the main Analytics solutions today) for your learning, completely free.

Software – Regardless of the programming language you choose, you will need to install the interpreter and an IDE. If your choice is R, for example, in addition to installing the language, you can install RStudio. The same goes for other programming languages. It is possible to create your Data Science scripts 100% online, via browser, using Jupyter Notebook. But you may not always be online, and having your tools installed locally will bring a series of advantages. In addition, consider installing:

– Text editors: Sublime, Atom, Notepad++
– Virtual Machine Software: VirtualBox
– Git and GitHub: to create your Data Science project portfolio
– Office Suite: Microsoft Office, LibreOffice

With the exception of Microsoft Office, all other tools are free.

Step 3: Statistics and Mathematics

Knowledge of Statistics and Mathematics is part of the essential package for anyone who intends to work as a Data Scientist. Statistical models and Machine Learning algorithms depend on knowledge of linear regression, multiple regression, clustering, Linear Algebra, etc. Do you need to be an expert in Statistics or Mathematics or even have a degree in these areas? The answer is no. Although these areas allow for a more comprehensive understanding, it is possible to learn these concepts and apply them throughout your Data Science learning journey. You do not need to learn every topic related to Statistics or Mathematics.

There are many ways to learn the concepts of Statistics and applied Mathematics, and this takes time. For any aspiring Data Scientist, the recommendation is to learn statistics by coding, preferably in Python or R, so that you can immediately apply a learned concept. Nothing replaces a degree in Statistics or Mathematics, of course, but you can learn the concepts that will be used in your daily Data Science work by applying these concepts through a programming language. Data Science is a multidisciplinary field.

At the end of this article, you will find some resources recommended by our team of experts!Analytics

Step 4: Big Data

Big Data is the raw material of Data Science. The Data Scientist profession arose from the need to create new methods of analyzing the immense volume of data that has been growing exponentially. Analytical techniques have existed for many decades (perhaps centuries), but never in human history has so much data been generated as today. New ways of collecting, storing, and analyzing data are needed, and Big Data is revolutionizing today's world because, with so much data at our disposal, we can make decisions in real time, and this directly impacts all of our lives.

The Data Scientist will consume Big Data, meaning they will use Big Data as raw material, apply various techniques, and gather insights. However, the responsibility for collecting and storing data usually lies with the Data Engineer. Creating Hadoop clusters, data streaming with Spark, and integrating different data sources are all new tasks typically performed by Data Engineers. But it is important for the Data Scientist to know well how the infrastructure that stores the data to be analyzed works, as this can make a difference when analyzing 1 trillion records, for example.

Hadoop – Hadoop is becoming the heart of Big Data infrastructure, which will revolutionize the traditional database storage system as we know it today. Besides being free, Hadoop was designed to run on low-cost hardware, an essential combination for companies looking to reduce their IT infrastructure costs and still capitalize on the benefits of Big Data.

Spark – Spark is an open-source project maintained by a developer community that was created in 2009 at the University of California, Berkeley. Spark was conceived with the primary goal of being fast, both in query and algorithm processing, in addition to in-memory processing and efficient fault recovery. It is currently one of the hottest topics in Data Science and has been gaining a lot of popularity.

NoSQL Databases – Traditional RDBMS (Relational Database Management Systems) databases were not designed to handle large amounts of data (Big Data). Traditional databases were designed only to handle datasets that can be stored in rows and columns and therefore can be queried using SQL (Structured Query Language). Relational databases are not capable of handling unstructured or semi-structured data. In other words, relational databases simply do not have the necessary features to meet the requirements of Big Data, where data is generated in large volumes and at high speed. This is the gap filled by NoSQL databases, such as MongoDB, for example. NoSQL databases are distributed, non-relational databases designed to meet the requirements of this new world of data we live in.

Relational Databases and Data Warehouses – In recent decades, all corporate data has been stored in relational databases, and Business Intelligence solutions have used Data Warehouses to create analytical solutions. This structured data will be the data source for Data Science, hence the importance of knowledge in SQL, the standard language for querying these types of data.

As a Data Scientist, do you need to be an expert in all technologies? No. But part of the Data Scientist's job will be to collect data from HDFS (Hadoop File System), create RDDs in Spark, apply Machine Learning algorithms to streaming data, cross-reference unstructured data collected from social networks with CRM databases, etc. Therefore, the Data Scientist needs to be comfortable with how data is stored and extract the best that technology has to offer.

Step 5: Programming Language and Machine Learning

There are several analysis tools, and the number of solutions keeps growing. But the recommendation for beginners is to obtain basic knowledge before trying to use analysis tools or commercial packages, in order to extract the best from these tools. Some programming languages have become icons in Data Science, such as Python and R, for several reasons: they are free, have an active and growing community, have already passed their maturation period, are widely used in both academia and business, and specialize in Data Science.

Python – It is a general-purpose language that has received more and more modules and packages for Data Science in recent years, such as Pandas, Matplotlib, Scikit-Learn, and Statsmodels. Python is easier to learn compared to other languages, has an active community, plenty of documentation available (including in Portuguese), and can be used for other activities besides Data Science.

R Language – A statistical language that has existed for over 30 years. Its ability to process statistics from large volumes of data and create sophisticated graphics has led technology giants like Oracle and Microsoft to adopt R as the standard language for statistical analysis. One of the most common problems people face when learning R is the lack of a guide. People do not know where to start, how to proceed, or which path to follow. There is an overload of good free resources available on the Internet, which makes the learning path much more winding.

For example: Microsoft's machine learning platform (Microsoft Azure Machine Learning) has a series of Python and R modules ready for use. Knowledge of at least one of these languages is fundamental.

Other languages like Julia, Scala, and Java are also widely used in Data Science, but if you are starting out, choose R or Python. These languages will provide a solid foundation, allowing the professional to advance to commercial solutions such as SAS, Microsoft Azure Machine Learning, Oracle Advanced Analytics, MicroStrategy, SAP Predictive Analytics, Tibco Analytics, among others.

If you already have knowledge of Matlab, Octave, Stata, or Minitab, know that your knowledge can already be used in Data Science.

An important tip: do not try to learn everything! Select 2 or 3 tools and dedicate yourself to learning them in detail.

And of course, you cannot talk about Data Science without talking about Machine Learning.

Machine Learning is one of today's most fascinating technologies. You probably use learning algorithms several times a day without knowing it. Whenever you use a search engine like "Google" or "Bing", one of the reasons they work so well is a learning algorithm. An algorithm implemented by "Google" learned to classify web pages. Every time you use the app to "tag" people in photos on "Facebook" and it recognizes your friends' photos, that is also Machine Learning. Every time your email's spam filter filters out tons of unwanted messages, that is also a learning algorithm.

Some reasons for the growth of Machine Learning are the growth of the web and automation. This means we have larger datasets than ever before. For example, many companies are collecting web click data, also called "clickstream" data, and are creating algorithms to mine this data and generate recommendation systems that "learn" about users and offer products they are most likely looking for. Netflix is one of the most successful examples of Machine Learning application. Every time you watch a movie or rate it, the system "learns" your taste and starts offering movies in a personalized way for each user.

There are several machine learning algorithms, depending on whether the learning is supervised or unsupervised, such as: Linear Regression, Ordinary Least Squares Regression (OLSR), Logistic Regression, Classification and Regression Tree (CART), Naive Bayes, Gaussian Naive Bayes, k-Nearest Neighbour (kNN), k-Means, Bootstrapped Aggregation (Bagging), Natural Language Processing (NLP), Principal Component Analysis (PCA), Principal Component Regression (PCR), Back-Propagation, and much more.

Each algorithm will be ideal for a certain type of data and according to the intended analysis. There is no need to learn all algorithms. But it is important to understand the concepts and how to implement them. Hence the importance of Mathematics and Statistics. Check out the tips at the end of the article.

Step 6: Business Knowledge

What is the goal of your analysis? Why are you going to collect mountains of data and apply analysis models? What problem do you intend to solve by analyzing data? The main goal of Data Science is to solve problems. Companies will not start a Data Science project if it is not relevant to the business. Therefore, the Data Scientist must be familiar with the business area for which they are starting a project using Data Science.

The market typically misunderstands this requirement for the Data Scientist profession, leading people to believe that the professional needs to be an expert in a specific business segment. But here it is worth making some important considerations. First, Business Intelligence professionals have always had this requirement: to know a business area well in order to collect KPIs (indicators) and thereby provide BI solutions that meet the client's needs. This has not changed; what has changed is how the analysis is done, now that Big Data has entered the equation. Second, given the scope of Data Science and Big Data projects, there will rarely be a single professional working, but rather a Data Science team, usually led by the Data Scientist. And in the Data Science Team, different profiles will act, such as specialists in business segments.

The tip here is simple. Try to understand the business area in which you intend to work as a Data Scientist. If you are going to work in a mining company, for example, what are the main indicators? Where does the data come from? What problems does the company need to solve? What types of data should be analyzed and correlated? How can Machine Learning techniques be used to improve the company's revenue? How does data analysis allow offering a better service to customers? Each business area has its particularities, and a broad understanding of this will allow for work that truly generates value.

Step 7: Presentation and Data Visualization Techniques

With the massive amount of data increasing every day, a major challenge is emerging for those responsible for analyzing, summarizing, and presenting data: making the generated information easily understood.

And one of the most important tasks of the Data Scientist's work is to be able to convey everything that the data wants to say. And sometimes data means different things to different audiences. It may seem easy at first. Today we have the most varied resources for presentation at our disposal, and that is exactly where the challenge lies. It has never been easier to generate tables and charts with different structures, formats, sizes, colors, and fonts. Charts are ceasing to be just charts and are becoming infographics. Having an ever-increasing volume of data at our disposal does not make presenting the generated information easier. On the contrary, it makes the task more complicated. Almost an art.

One of the Data Scientist's tasks is to present their results. No one is better than the professional who performs the analysis, from collection, cleaning, and storage of data to the application of statistical models, to explain their results. Effective data visualization can be the difference between success and failure in business decisions. Personally, I believe that soon, the ability to communicate and tell stories with data will be one of the most valued and sought-after characteristics by companies. Presentation techniques are one of the skills that will make a difference when hiring a Data Scientist, as technical knowledge becomes widespread.

The Data Scientist must be a storyteller and must be able to tell the same story in different ways. The professional who is able to combine the technical skills required for data analysis with the ability to tell stories will be a unique professional. This is the true concept of a unicorn, attributed to Data Scientists.

Several tools have advanced features for data visualization: Pentaho, Tableau, QlikView, Microsoft Excel, Microsoft Power BI, MicroStrategy, Weka, NetworkX, Gephi, JavaScript libraries (D3.js, Chart.js, Dygraphs), as well as high-level visualizations that can be done in Python or R.

It is important not only to be familiar with one or more tools to visualize data, but also the principles behind the visual encoding of data and communication of information.

Step 8: Practice!!

There is no other way to learn whatever the subject is. You need to practice, test, experiment, make mistakes, learn from them, test again, and interact with the community.

Phew. If you've made it this far, congratulations! Now do you better understand why Data Scientist is a rising profession and why its compensation is among the highest in any survey conducted? But we're not done yet, keep reading:

Errors to avoid along the way:

1- Thinking that learning is easy and fast – In no profession is learning fast and easy. Acquiring knowledge and experience requires time, effort, investment, and a lot of dedication. Do not fall into this trap. Start your preparation today, but be aware that acquiring knowledge takes time.

2- Learning too many concepts at the same time – This is the most common mistake. Because it is an interdisciplinary field, there is usually a tendency to want to learn too much at once. Do not focus on quantity, but rather on quality. Learn a concept, consolidate it, practice, and only then move on to another area of study. The beginning will be harder, but as you learn and gain experience, learning new tools becomes easier.

3- Starting with very complex problems – Solving more complex problems in Data Science requires time and experience. Do not try to do this at the beginning of your journey.

4- Focusing only on programming – Data Science is not just programming. Other concepts are just as important. Statistics is important. Data visualization and presentation are important. Keep your focus on other areas and not just programming.

And how to study?

Before investing in analytical training, take these steps to make sure you get real value from your investment.

1. Define your goals. As the saying goes: "If you don't know where you're going, any road will take you there!". Did you do the self-assessment suggested at the beginning of the text? What was the result? What are the areas you need to dedicate yourself to? What will your availability be? What is your goal? I know there are many questions, but they need to be answered before defining your learning path.

2. Participate in communities. Data Scientists usually meet at Meetups and on some blogs (links below). These communities will allow for the exchange of experiences, and this is very valuable.

3. Experiment. Until you run a machine learning algorithm on a dataset with millions of records, create a data cleaning and transformation algorithm, or collect streaming data from social networks, you won't understand how things work. Experiment! Learn, do, make mistakes, do it again, and when you least expect it, you will be able to analyze data and contribute to the company where you work or to your own business.

Use the information you gathered to select training options that help you achieve your goals without wasting time and money.

Knowledge in Data Science requires time and dedication. The training you choose should be a mix of theoretical foundation, practice, and experimentation.

Resources

There are many resources available to learn Data Science, and so many resources end up causing overload for beginners, who can lose focus. We have separated here a list of what we consider the best sources of learning in Data Science.

Blogs:

Data Science Central: http://www.datasciencecentral.com
KDnuggets: http://www.kdnuggets.com
Articles about R: http://www.r-bloggers.com
Python Brasil: http://python.org.br

Statistics:

Statistics: http://www.statistics.com
Simply Statistics: http://simplystatistics.org

Machine Learning:

Machine Learning Coursera: https://www.coursera.org/learn/machine-learning
Deep Learning: http://deeplearning.net/
Deep Learning Book: http://www.deeplearningbook.com.br

Videos and Competitions:

Top 10 TED Talks for Data Scientists: http://www.kdnuggets.com/2016/02/top-10-tedtalks-data-scientists.html
Data Science for Social Good: http://dssg.uchicago.edu
Kaggle: https://www.kaggle.com
Data Science Game: http://www.datasciencegame.com

Training:

Data Science from Harvard: http://cs109.github.io/2014
Data Visualization: https://columbiadatascience.com/category/course-topics/data-visualization
Open Data Science Master: http://datasciencemasters.org
Big Data and Social Analytics MIT: http://getsmarter.mit.edu/big-data-and-social-analytics-course-aw
Metis Data Science: http://www.thisismetis.com
Zipfian Academy: http://www.zipfianacademy.com
e-Setorial Business Analytics: http://www.e-setorial.com.br/servicos