Learning Data Science does not start with doing Data Science

Those who are in love with practice without theoretical knowledge are like the sailor who goes onto a ship without rudder or compass and who never can be certain whither he is going. Pra

The foundation of Data science lies on a set of math and statistics knowledge which provides the ability to understand and model datasets. Because data modeling algorithms may look to the untrained eye like simply a collection of lines of code, those jumping into modeling data without the statistical or mathematical background, may end up making serious mistakes that would nullify the results of any analysis done.

If you are serious about becoming a Data Scientist, avoid the temptation of jumping into modeling data first. Instead, ensure you have the right math and statistical foundation in place before learning how to build models. For details on the overall skills required to be a Data Scientist, read my previous blog What are the skills that define the role of Data Scientist? For resources on alternatives to become a Data Scientist, read Where do Data Scientists come from?

There are multiple ways to get skills to become a Data Scientist, but independently of the option you choose, make sure the right foundation is built into the program, or find ways to build the foundation yourself either through college classes, digital classes, or even through good old fashion academic books. Specifically, what math knowledge should you have? here is a summary of some of the most common math knowledge needed:

Probability Foundation

What is Probability?, Sample Spaces, Properties/Rules of Probability, Probability of combination of events (Intersection of events, Union of events, Contingency Tables), Conditional Probabilities, Independent vs. Dependent Events, Bayes’ Theorem, Counting Principles (permutations, Combinations), Sampling Techniques, Probability Distributions (Continuous and Discrete Random Variables, Cumulative Distribution, Binomial Distribution, Poisson Distribution, Geometric Distribution, Exponential Distribution, Normal Distribution, Chi-Square Distribution, Expected Values)

Statistics Foundation

  • Descriptive Statistics: Quantitative Data vs. Qualitative Data, Types of measurements: nominal, ordinal, intervals, ratios, Frequency Distributions, Relative Frequency Distributions, Cumulative Frequency Distributions, Measures of Central Tendency: Mean, Median and Mode, Measures of Dispersion: Range, Variance, Standard Deviation, Measures of Relative Position: Quartiles, Interquartile Range, Outliers, The empirical rule (normal distributions) and Chebyshev’s Theorem, Visualizing Data: Histograms, Stem and Leaf, Box Plots
  • Inferential Statistics: Sampling distribution of the mean, Sampling Distribution of the Proportion, Standard Error of the Mean, The Central Limit Theorem, Confidence Intervals and their interpretation, Effects of changing confidence levels and sample sizes, Working with small vs large samples, Formulating and testing hypothesis, The Null and Alternative Hypothesis, Type I and Type II Errors, 1-tail vs 2-tail hypothesis, Testing the Mean and the proportion of a population using 1 sample, Testing the difference in means and proportions using 2 samples, Analysis of Variance (ANOVA) comparing 3 or more population means, Understanding the role of Alpha and the p-value, Working with Dependent vs. Independent samples, Correlation and Simple Regression (confidence intervals, hypothesis test on the regression line, regression assumptions), Multiple Regression (assumptions, multicollinearity)

Linear Algebra Foundation: Vectors and Real Coordinate Spaces (Adding, Subtracting, and multiplying vectors by a scalar, Linear Combinations, Linear dependence and independence), Matrix Transformations (Functions and Linear Transformations, Inverses and determinants, Matrix Multiplications, Transpose), Orthogonal Complements and Projections, eigenvectors, eigenvalues.

The bottom line is that when you move from working in a deterministic world to probabilistic world, it is important to understand the implications of this paradigm change.

Where do Data Scientists Come From? How Do I become One?

interior design-2

I frequently get asked…. How do I become a data scientist? How do Data Scientists get their skills? There are several options but first… Let’s take a look at the profile of data scientists and see how your skills rank compared to them.  According to a 2018 study by Burtch Works, it is estimated that:

  • 91% of of data scientists have an advanced degree (43% hold a Master’s degree, and 48% hold a PhD).
  • 25% of data scientists hold a degree in statistics or mathematics, 20% have a computer science degree, 20% hold a degree in the natural sciences, 18% hold an engineering degree, 8% hold a business degree, 5% hold a social science degree, and 4% an economics degree
  • 44% of data scientists are employed by the technology industry, followed by Financial Services with 14%, marketing services with 9%, consulting 8%, Healthcare/Pharma 6% and retail with 5%
  • 62% of data scientists are US citizens, 19% have permanent residency and the rest are on temporary visas
  • 85% of data scientists are male

These statistics should give you a general idea about the skillset shared by the community of data scientists.

How do I become a Data Scientist? Where do I get the skills?

There are several options depending on how much time you want to spend acquiring the skills, how much money you want to spend, and how deep you want your skills to be. Here is a summary of the most popular options:

1-PhD: If you are interested in a PhD, or already pursuing a PhD, most likely than not, you will acquire skills that will help you become a Data Scientist. As we saw in the statistics above, today almost half of data scientists have a PhD.  On average, a PhD will take several years to complete.

2- Master of Science in Analytics/Data Science: The first master program focused on Analytics/Data Science was started by the Institute of Advanced Analytics (IAA) at NC State in 2007 (I graduated from this program in 2012). It was for many years the only program in the nation, and due to high demand, many new programs have been launched in the past few years. There are now over 200 programs in the United States, with some of them full time, some part time, and some online.  If you are curious about what programs are available that could match your needs, you can use the Interactive Map kept by the IAA to find options that might fit your needs in the United States. It will usually take you 1 or 2 years to complete a master degree.

3- Data Science Bootcamps: There are several alternatives.  Switchup has a good summary of the best Data Science Bootcamps.  Figure 1 summarizes a comparison of these programs using Switchup’s ratings and details from each of the programs (Disclaimer: this information varies frequently so this may be out of date by the time you read it). Refer to their web sites for up to date details. You will be able to complete a bootcamp in a matter of months.

Figure 1 Summary of Information on the Best Data Science Bootcamps (Switchup)

4- Data Science Online Certifications. MOOCs provide a a wealth of training options: Coursera, Edx, Cognitiveclass.AI, and Udacity are all great choices. Udacity has an advantage over the others in that they offer mentorship together with their education programs. (See what one student of both Udacity and other programs has to say about his experience). Prices and length of programs vary greatly among these choices, and they offer a lot of flexibility.  However, you will still need practical experience.  These classes can give you a foundation, that will need to be supplemented with practical experience to become a Junior Data Scientist.

For more information on the type of skills needed to be a Data Scientist, refer to What are the skills that define the role of Data Scientists?

What are the skills that define the role of Data Scientists?

Data Science is an emerging field, but it is definitely not a new field. Yet, many people still struggle to define Data Science as a field, and more importantly, struggle to define the set of skills that collectively define a “Data Scientist”.

What is data science?

Data Science is a cross-disciplinary set of skills found at the intersection of statistics, computer programming, and domain expertise.  Perhaps one of the simplest definitions is illustrated by Drew Conway’s Data Science Venn Diagram (Figure 1), first published on his blog in September 2010.  Discussions about this field, however, go as far back as 50 years.  If you are interested in learning more about the history of the Data Science field, you can read it in the  50 Years of Data Science paper written by David Donoho.

Figure 1 – Drew Conway’s Data Science Venn Diagram

The bottom line is that Data science comprises three distinct and overlapping areas: a set of math and statistics knowledge which provides the ability to understand and model datasets, a set of computer programming/hacking skills to leverage algorithms that can analyze and visualize data, and the domain expertise needed to ask the right questions, and put the answers in the right context.

It is important to call out attention to the “Danger Zone” above, as there is nothing more dangerous than aspiring Data Scientists who do not have the appropriate math and statistical foundation to model data.

What skills define the role of Data Scientists?

A Data Scientist is not a just a computer programmer, or just a statistician, or just a business analyst.  In order to be a data scientist, individuals need to acquire knowledge from all these disciplines, and at the minimum develop skills in the following areas:

1.Probability, Statistics, and Math foundation. This includes probability theory, sampling, probability distributions, descriptive statistics (measures of central tendency and dispersion, etc.), inferential statistics (correlations, regressions, central limit theory, confidence intervals, development and testing of hypothesis, etc.) and linear algebra (working with vectors and matrices, eigenvectors, eigenvalues, etc.)

2.Computer Programming.  Throughout the years, SAS has probably been the most commonly used programming language for Data Science, but adoption of Open Source Languages Python and R has increased significantly.  If you are starting today to acquire data science skills, my recommendation would be to focus on Python. Looking at worldwide searches on Google for “R Data Science” and comparing them to “Python Data Science”, the trends are clear (Figure 2). Interest in Python has surpassed R, and continue on a positive trend. This makes sense given that python allows you to create models and also to deploy them as part of an enterprise application, so within the same platform data scientists and app developers can work together to build and deploy end to end models. R while easier in some cases for modeling purposes, was not designed as a multi-purpose programming language.

Figure 2 Worldwide searches for “R Data Science” vs. “Python Data Science”. Google Trends (June 2018)

3. Data Science Foundation. This involves learning what data science is and its value in specific use cases. It also involves learning how to formulate problems as research questions with associated hypotheses, and applying the scientific method to business problems.  Data Science is an iterative process so it is critical to have a solid understanding of the methodologies used in the execution of this iterative process (Define the problem, Gather Information, Form hypothesis, Find/Collect data, Clean/Transform data, Analyze Data and Interpret Results, Form new hypothesis)

Figure 3 Data Science Iterative Cycle

4. Data Preparation/Data Wrangling.  Data is by definition dirty.  And before data can be analyzed and modeled, it needs to be collected, integrated, cleaned, manipulated and transformed. Although this is the domain of “Data Engineers”, Data Scientists should also have a solid understand of how to construct usable, clean datasets

5. Model Building. This is the core of the data science execution, where different algorithms are used to train models with data( structured and unstructured) and the best algorithm is selected.  At this stage, data scientists need to make basic decisions around the data such as how to deal with missing values, outliers, unbalanced data, multicollinearity, etc. They need to have solid knowledge of feature selection techniques (which data to include in the analysis), and be proficient in the use of techniques for dimensionality reduction such as principal component analysis. Data scientists will be able to test different supervised and unsupervised algorithms such as regressions, logistic regressions, decision trees, boosting, random forest, Support Vector Machines, association rules, classification, clustering, neural networks, time series, survival analysis, etc. Once different algorithms are tested, the “best” algorithm is selected using different model accuracy metrics. Data scientists should also be skilled in data visualization techniques, and should have solid communication skills to properly share the results of the analysis and the recommendations with nontechnical audiences.

6. Model deployment.  A very important part of building models is to understand how to deploy those models for consumption from a data application. While this is typically the domain of machine learning engineers and application developers, data scientists should be familiar with the different methods to deploy models.

7. Big Data Foundation. A lot of organizations have deployed big data infrastructure such as Hadoop and Spark.  It is important for data scientists to know how to work with these environments.

8. Soft Skills.  Successful data scientists should also have the following soft skills:

a.    Ability to work in teams. Because of the inter-disciplinary nature of this field, it is by definition a team sport.  While every data scientist on a team will need a good foundation on all skills defined above, the depth of skills will vary among them. This is not a field for individualistic stars, but a field for natural team players.

b.     Communication Skills. Data scientists need to be able to explain the results of their analysis and the implications of those results in nontechnical terms. The best analysis can go to waste is not properly communicated.

Last but not least, it is important to remember that the most important characteristic of great data scientists is CURIOSITY. Data Scientists should be relentless in their search for the best data and the best algorithm, and should also be lifelong learners as this field is advancing very rapidly.

In summary, if you are interested in the Data Science field, or if you are exploring ways to develop your skills, make sure that you are addressing all these areas, and especially make sure not to end up in the danger zone having programming skills and domain knowledge but lacking the math and statistics foundation needed to model data correctly.

What Data Science Tools Have you used in the past 12 months? KDnuggets Poll Results are out!

The results of the 18th annual KDnuggets Software Poll were recently published. This poll asks “What Predictive Analytics, Data Mining, Data Science Software/Tools have you used in the past 12 months?”.  This poll attracted 2,900 voters, and it is also worth mentioning that it sometimes attracts controversy due to excessive voting by some vendors.  See all data at KDNuggets.

Some of the most relevant findings are:

  • Python has now overtaken R as a Data Science Tool – barely but still noticeable (53% to 52% use but Python grew 15% while R only grew 6%)
  • There are now 2 newcomers joining the top 10 list: Tensorflow and Anaconda
  • Use of Excel for Analytics purposes decreased by 16%
  • In terms of programming languages, Python, R and SQL run the show with usage of all 3 growing
  • Big Data Tools was simplified to only 4 categories: Hadoop Open Source, Hadoop Commercial, SQL on Hadoop Tools and Spark.  The highest growth tool is SQL on Hadoop and usage of Hadoop Open Source is decreasing

TOP SOFTWARE TOOLS

We have 2 newcomers this year: Anaconda and Tensorflow

Top 2 tools:

  • Use: Python (53%) and R (52%)
  • Growth: Tensorflow (197%), Anaconda (37%)

TOP LANGUAGES

Top 2 languages:

  • Use: Python (53%) and R (52%)
  • Growth: Python (15%), R (6%)

TOP BIG DATA TOOLS

screen-shot-2017-05-24-at-10-46-35-am.png

The tools on the survey have been simplified to 4: Hadoop Open Source, Hadoop Commercial, Spark and SQL on Hadoop Tools.

Top 2 Big data tools:

  • Use: Spark (23%) and Hadoop Open Source (15%)
  • Growth: SQL on Hadoop Tools (41%), Spark (5%)

It is important to notice the decrease of 32% in usage of Hadoop Open Source. I am not sure if there has been a real decrease, or if this is the result of the survey having changed splitting the hadoop category in 2: Open Source and Commercial.  Part of this “decrease” could be attributed to the fact that there are now 2 categories instead of 1.

DEEP LEARNING TOOLS

Top Deep Learning Tools:

  • Use: Tensorflow (20%) and Keras (9.5%)
  • Growth: Microsoft CNTK (278%), mxnet (200%)

What Data Science Tool Have you used in the past 12 months?

The results of the 16th annual KDnuggets Software Poll were recently published. This poll asks “What Predictive Analytics, Data Mining, Data Science Software/Tools have you used in the past 12 months?”.  This poll attracted 2,800 voters, and it is also worth mentioning that it sometimes attracts controversy due to excessive voting by some vendors.

93 different tools were included in the poll. To determine which tools are included, they start with the companies in Gartner Magic Quadrant(tm) for Advanced Analytics and Forrester Wave(tm) for Big Data Predictive Analytics, and add companies/tools from last year poll, and relevant new ones in the market.

R is the tool most frequently used tool in this community of data scientists, but other tools are growing rapidly (Spark, KNIME, Python).

Top 10 data mining tools

 

In terms of programming languages, Python is the clear leader.

kdnuggetslanguages

 

And Hadoop is the clear leader in the Big Data Tools space.

big data tools