Georgia Tech OMSA : My Program Review

Updated 5/12/2020: MGT 8803 — Business Fundamentals for Analytics, ISYE 6748: Analytics Practicum (Graduated with 3.88 GPA)

Updated 12/14/2020: ISYE 6740 — Computation Data Analytics, ISYE 6644 Simulation

Updated 08/05/2020: ISYE 8803 — Special Topics in High Dimensional Analytics (GPA still 4.0)

tl,dr: This program was worth every cent. Since being a part of this experience, I have not only grown exponentially in my field, but I’ve been published, I’ve been a presenter at Nvidia’s GTC conference, I’ve become a mentor for aData Science bootcamp. I’ve done so much more than I ever could have fathom the day I first submitted my application.

I graduated Spring of 2021 from the Master’s of Analytics from Georgia Tech, and I wanted to share my story, experience, and reviews. I officially enrolled in Fall of 2019 and completed the program in 1.5 years. There was excitement, there were highs (and lows), a few tears, a lot of sleepless nights, but overall it was a great experience that honestly, I was sad to have it come to an end. Mostly because the program grows every semester. Each term they add more courses to the catalog, and there’s easily 9 or 10 of them that I just wish I had the opportunity to take during my time in the program. But I’m here to talk about my story and experience, as well as answer any lingering questions you may have about the program.

The first question I always get from anyone interested in the program is “how hard is it and how hard is it to get in?”. Well, unfortunately there’s no answer to this that will suit every person. There’s undoubtedly high correlation between school rankings and admission standards / curriculum rigor. So let’s put this in perspective. The Masters in Analytics is an interdisciplinary degree between the College of Engineering’s Stewart School of Industrial & Systems Engineering, which is currently the #1 school according to U.S. News

Scheller College of Business, which is currently the #27 Business school, but their Business Analytics Program is #3 (All your courses will be Business Analytics)

and The college of Computing’s School of Computational Science and Engineering, which Georgia Tech is ranked #8 for computer Science and #7 for AI. Both of these you will be focusing more on than Systems and Theory.

So, it’s safe to assume it won’t be a walk in the park. And I can unequivocally confirm this. The only answer I have for the question at hand is

Based off of your own experience, mileage may vary

The truth is that the program is full of diverse people from extremely diverse backgrounds.You can read reviews on almost all courses at omscentral, but keep in mind that the reviews may be coming from someone with 20+ years of software engineering experience to someone from a purely financial background who just started learning Data Science from MOOCs on Coursera. So, keeping all things even, take what I say here with the notion that my background is…….

My Background

I received my Bachelor’s in Physics from Ole Miss back in 2010 where I graduated with a 3.25 GPA. The curriculum obviously was very math intensive, but we also had to take computer programming in Fortran and C++. I did very well in both of these courses, but I definitely did not become a software engineer. After graduating, I got a job as a petroleum engineer doing consulting on oil and gas rigs. During this time I did little to no computer programming, but there was some analytics involved (nothing to the extent of a Data Analyst or Data Scientist). I kept up a little with my C++ skills, but even today I wouldn’t consider myself a C++ programmer.

It was around 2015 that I took my first Machine Learning Course. As you can probably guess, it was Stanford’s ML course on Coursera. If you haven’t taken this yet but are considering the OMSA, do yourself a favor and stop reading this and go. Here is the link. After completing the 12 week course in 3 weeks (it is that good), I continued on with the Python For Everybody Specialization as well as the Applied Data Science in Python Specialization. Both offered by the University of Michigan. It was in these courses that I really refined my Python skills. I took these courses while still performing my normal job. I took a few other MOOCs on Coursera, but let’s be honest, MOOCs are not sufficient enough on their own to make a complete career switch into Data Science or the Big Data world. However, here is a short list of some of the courses I took

• Statistics with Python
• Advance Data Science with IBM Specialization
• Deep Learning Specialization
• Complete Guide to TensorFlow for Deep Learning with Python
• Spark and Python for Big Data with PySpark

I also did some Kaggles, but to be honest it wasn’t really my thing. I felt that there was nothing original about working on the same datasets that millions of other people had already explored millions of times. I was looking for something more original and challenging. That’s when I joined Springboard’s Data Science Career Track program. It is a 6 month bootcamp with a job guarantee at the end. What really allured me to the program was the fact that they paired you with a mentor, someone who is a practicing DS in industry, as well as a full career staff behind you.

Upon completion of the bootcamp, I not only switched over to a Data Scientist role with my company, but they liked my work so much that I’m actually featured on their website. You can check out my github repo from the program to get an idea of what you would be doing if you choose to join.

APPLYING

So in full disclosure, I had to apply to the program twice. The first time I applied was summer 2018 and I was late to the application game. By the time I got my application in and all three letters of recommendation submitted, it was almost the deadline for normal admission. Due to the shear amount of applicants the program gets, of course I was rejected. However it wasn’t pure rejection as there was an extra “your application caught our eye” notice. After getting in touch with admissions, they advised me to apply much earlier, as well as enroll in the Micro Masters Program. This turned out to be great advice because of two reasons:

1. I was accepted into the program
2. The MM program allows you to earn credit for up to 3 classes in the full Master’s program. I received over 90% in the MM and entered into the OMSA program with credit for 3 courses. This alone reduced my time commitment in the full program by at least 2 semesters.

The program has 3 different tracks:

• Computational
• Business Analytics
• Analytic Tools

I elected for the Computational track which is basically all graduate level Computer Science courses with some business courses sprinkled in. The majority of my courses are with the OMSCS students

COURSES

Note: These are my opinions and reviews based off of my experience. To assure you that there is no hidden bias in anything negative said about a course, please note that I graduated with a 3.88 GPA. So negative comments are due to my true feelings about a course, not because I did poorly in a particular class. No sour grapes here.

In the MM program you take courses with students in the OMSA and it is the actual course from the program. Because of that, my reviews will include the Micro Masters’ courses. These will be designated with a * next to them. I’ll prefix each course with an index number. Courses with the same number means I took them concurrently. I’ll try and make a note on how taking two classes at once might have impacted my overall performance.

1 — ISYE 6501: INTRODUCTION TO ANALYTICS MODELING *

Difficulty: 3/5

Enjoyment: 4/5

Time Commitment: 8 hrs/week

This course is more or less a survey course over a multitude of analytical processes and models. It covers everything (not in great detail) from Machine Learning and Time Series to Simulation and Optimization techniques. The homeworks weren’t very challenging (a lot of the MOOCs I had previously taken already covered some of the material — so I came into this course with a solid foundation) and they were peer graded. This would cause a false sense of accomplishment to be honest. It was nothing to get perfect scores on the HW, but the exams were sufficiently challenging. There were 2 midterms and a final that expected much firmer grasp of the material than what the homeworks required. The lectures were well organized and I felt the deliverance of the material was adequate. I do enjoy Dr. Sokol and his presentation style. The course is primarily in R with the simulation and optimization sections using Python. The simulation challenge had to do with figuring out the optimal amount of attendants at the check-in counter for the atlanta airport so as to keep wait times to a certain goal. The optimization problem was a derivative of the Stigler Diet problem. Overall I really enjoyed this course. It’s not too difficult to get an A regardless of your background.

1 — CSE 6040: COMPUTING FOR DATA ANALYSIS *

Difficulty: 4/5

Enjoyment: 5/5

Time Commitment: up to 20 hrs/week

This course should be rebranded as “Algorithms in Python — on steroids”. This course was very challenging, and I could easily see it as a weed-out course to weed out those students who probably didn’t belong in the program. As stated, we only used Python and it introduced a user to all the different ways to use Python to interact with data. Homeworks utilized Jupyter Notebooks with autograders and you got 2 weeks to complete them. Some of them absolutely took the full two weeks. This was a very algorithmic course in which you were tasked in constructing unique algorithms to solve problems. In some of the questions, it wasn’t enough to solve the problem, but it had to be an efficient solution. Those particular questions were timed during execution, and if run time was longer than a certain threshold, you failed. We covered things like using Python to interact with SQL databases, the Pandas library for data manipulation, Singular Value Decomposition, as well as building a recommender system from scratch. There were 2 midterms and a final, all of which were INTENSE. They also are in Jupyter Notebook format as well as open notes and open internet (just not open collaboration). They allow this because each midterm is made just for that semester and then never used again. So it’s not like you could just google the answer. We got 24 hours to complete the first midterm, then 36 hours to complete the 2nd midterm as well as the final. It took me about 8 hours to complete the first, 12 hours to complete the second, and I’m pretty sure I used almost all 36 hours to complete the final. They make them so that each question builds off the previous. So if you couldn’t solve the prior challenges, it is impossible to answer any of the later challenges. This forced you complete each step sequentially. Overall I loved this course for the challenge. At this point, I had 3 years of Python programming under my belt and this course still found ways to challenge me. Do not take this course if you have 0 Python experience. It is one thing to solve algorithms, it is another thing to solve algorithms in a language you have never used before.

2 — MGT 6203: Data Analytics for Business *

Difficulty: 1/5

Enjoyment: 2.5/5

Time Commitment: 2 hrs / week

I’ll be honest, I vaguely remember anything from this course. That is because I paired it with arguably the toughest class in the whole curriculum (see below this review). I do remember it starts off with in depth coverage of regression modelling and analysis. Regression was actually touched on twice in this course. Once in the ML realm of things and secondly in financial analysis — how to apply / interpret regression to stocks. That part was very interesting to me. The course covered broad topics from Finance, Marketing, and Supply Chain. The homeworks weren’t very challenging and often were word for word from the lectures. There were a few tricky questions in there from time to time, but nothing insurmountable. The midterms and finals were open notes. Overall, I have no business experience or background and spent minimal time on this course and was able to achieve 94%. I honestly believe I would’ve taken away more from this course if I was able to focus more on it, but again, I took it with a very challenging course that demanded all my time and effort.

2 — CSE 6250: Big Data for Healthcare

Difficulty: 5/5

Enjoyment: 5/5

Time Commitment: up to 30 hrs /week

This course is a beast. This is basically “The entire Hadoop and Spark Ecosystem in One Semester”. You get the environments in the form of a Docker container and run the gauntlet of Hadoop Pig, Hive, MapReduce, HDFS, Streaming as well as Spark, Spark SQL, Spark MLLib, and Spark GraphX. Fortunately the MapReduce section for Hadoop was coded in Python as opposed to Java (which I have little to no experience in), but the Spark sections were 100% Scala. Fortunately I had a solid PySpark background before taking this course and it wasn’t too hard to convert over to Scala Spark, but it was still a challenge. You’re practically learning a new language and ecosystem concurrently. To be quite frank, I watched maybe the first 2 weeks of lectures before ignoring them all together. The homeworks were each projects on their own and were very time consuming. The lectures offered very little in way completing the homeworks. It’s called Big Data for healthcare because all the datasets were EHR based (Electronic Health Records). This course does give you insight into becoming a Data Scientist / Machine Learning Engineer in the Healthcare field, but all the principles learned definitely extend outside of healthcare. It is assumed that students already have sufficient background in Machine Learning Principles, Deep Learning Principles, Linux, Command Line, and some distributed processes. If you are lacking in any of those, do not take this course. One of the homeworks had us calculating how many FLOPs were in a Convolutional Neural Network of specific architecture. If you don’t know what a CNN is or how it works, you’d be completely lost. It is not a machine learning course, it is a course on how to do machine learning at scale in distributed systems. There was 4 homeworks and a group project. My team decided to tackle the CheXpert competition hosted by Stanford where we used Spark on AWS EMR clusters to preprocess approx 460GBs of X-ray images into 2TBs of images and then used Tensorflow 2.0 to create a unique model architecture to make predictions. We unfortunately didn’t get to finish in time before the paper was due. We ended up training 14 intermediate models and one final concatenated model. One epoch of training time took 20 hours on a massive GPU instance on AWS and we were achieving 74% accuracy after only 2 Epochs. I plan on continuing the work we did to try and top the leaderboards. Our semester was the first in which the professor elected to have a final. From what I remember, it was pretty challenging, but not impossible. Overall I loved this class so much that I built my own Hadoop and Spark cluster out of raspberry pi 4’s. I used my 3D printer to construct the cluster case.

3 — CSE 6242: Data and Visual Analytics

Difficulty: 3/5

Enjoyment: 4/5

Time Commitment: up to 25 hrs/week

This was an interesting course. We covered things like utilizing API’s to scrape data and build social network graphs, build interactive data dashboards with D3.js, utilizing both Azure and AWS cloud services to perform MapReduce and Spark processing, build a Random Forest algorithm from scratch and construct the page rank algorithm in a distributed manner (no matrix). I’m not 100% sure what this class was supposed to be since it threw a little bit of everything at you, but overall I did enjoy it. There were 4 homeworks, which depending on your background, would seem like just busy work or a complete kick in the rear side for you. This one is one of the required core courses in the OMSA program, so everyone has to take it regardless of your background. Our class was over 1000 students and there were some serious complaints. If you knew what you were doing, you could knock out the HW in like a day or so. If you were on the other side of the spectrum, you would be living on piazza asking non stop questions for 2 weeks straight. It’s tough to prepare for this course if you haven’t had any real exposure to these concepts before because the class covers so much. Fortunately I work as a data scientist and perform a lot of the tasks covered on a daily basis. I knocked out each homework in about 2–5 days. The most interesting part for me was learning D3.js. There are some amazing websites out there using D3.js — check some out here — and it was great getting to learn how to replicate some of these sites. There was a group project that was very open ended. You could pick your own topic, data, and presentation method. The only hard requirement was that the data had to be of significant size (but no hard requirement on what that size was haha). Our team did a Scrollytelling website utilizing data on WW2 scraped for over a dozen sources. Unfortunately one of our team members dropped the course over halfway through and left us hanging. We didn’t get to complete what we had set out to do, but we still got enough done to get a 100 on the final project. I’m hosting the site from my own personal AWS site, you can check it out here. Overall, I enjoyed this class. You definitely get introduced to many techniques. I’m just a little confused on what the main purpose of this course was supposed to be. I tried not to spend too much time on this one because I paired it with Reinforcement learning, and let me tell you this — I haven’t been this miserable since running two-a-day football practices in 100 degree weather.

3 — CS 7642: Reinforcement Learning

Difficulty: 5/5

Enjoyment: 2.5/5

Time Commitment: 25 hours / week

I will start the review off on this one with, I probably would have enjoyed it more if I took this course by itself. But unfortunately I was glutton for punishment and paired it with CSE 6242. I will quote one of my fellow classmates on this course in describing it as “a marathon”. The amount of material and the depth in which the material is covered is just ridiculous. There is a lot of theory, proofs, convergence theorems, paper reading, and applications in this course. To be fully honest, in order to slightly understand everything going on, I had to supplement the lectures and readings with the lectures from David Silver of Google’s DeepMind (the guys who trained a reinforcement learning agent to play atari video games better than humans). There was 6 homeworks, 3 projects, and a final. The homeworks were somewhat challenging, but could be solved in a day. The first project was replicating results from an experiment of a published paper. This one was tough because there were so many details left out in the original paper on how to replicate the experiment. The slightest change in implementation, wildly changed results. The second experiment was a little more open ended, and fun. We had to solve the lunar lander problem from the OpenAI Gym environment. I utilized Deep Q Learning with Memory replay and Tensorflow 2.0 to train an agent to land on the moon. It wasn’t enough to just solve the problem however. You had to provide an in depth analysis on the execution of your agent’s actions based on the method used to solve the environment. The 3rd project broke me. Again, you had to replicate the results from a published paper. This project centered around Game theory and a soccer environment. You had to code a soccer game from scratch (you can see my implementation of the game below), and all the dynamics of the game in accordance to the rules as described in the original paper (which again were ambiguous). Then you had to train two competing agents to act in accordance of 4 different game theory approaches {Q-learning, Friend-Q, Foe-Q, and CEQ}. The main takeaway is the fact that Foe-Q and CEQ required an algorithm that utilized Linear Programming to optimize agent behavior. The course is taught by THE Charles Isbell and Michael Littman, a professor from Brown University, and their lectures were entertaining. It’s just the amount of material covered was borderline too much for one semester. I got a 100 on everything except the final which was true/false with explanation required. I got a 72, but the class average was 52 just to give you an idea of how tough it was. There is a massive curve for this course however. I believe from the stats I saw on slack and piazza that anyone who was at the class average or higher automatically got an A. Again, if I was taking this one by itself I probably would have enjoyed it more. RL is an amazing field with nowhere to go but up. I was amazed watching my little agent land on the moon by himself and how simple it was to get him to learn how to do it. And the guys at DeepMind are doing crazy things, they trained an agent to beat professional Starcraft players.

4 — ISYE 8803: Topics in High Dimensional Data Analytics

Difficulty: 3.5/5

Enjoyment: 5/5

Time Commitment: 15 hours / week

This is a class right here!! This is hands down the best class I’ve taken in the whole program thus far. Before I get into why, let me first get the negatives out of the way. In order to truly appreciate this class, you have to take it by itself so you can devote all your time and attention to, not just “getting” the material, but fully absorbing and understanding it. There’s two reasons for this. First— This a math and math notation heavy course. We aren’t talking just multiplications and additions, we are talking kroenecker and khatri rao products of matrices, N-way tensor decompositions, and closed form solutions to convex optimizations. If you don’t do this stuff or haven’t seen it at all, it will take a little bit for it to soak in. Second — the lectures don’t do the best job of explaining the math (in my opinion). For the most part, the professor just threw the equations and proofs up on the slide and did a lot of hand-wavy “as we can see, this goes to this….this means this….etc, etc”. The next negative that I have is that the response time from the TA’s was found to be lacking. This surprised me since on omscentral that was one of the biggest raves of the course, the activity level of the TAs and the professor. I think I only saw the professor pop up on piazza when a TA specifically directed a question to him that they couldn’t answer. Now on the TA part, I will say the response time could have something to do with the reshuffling of TAs that randomly happened at the beginning of the semester. But I don’t know for certain. The last complaint that I have is that the exams were horribly worded and ambiguous to the point that the majority of the class couldn’t even understand what some of the questions were asking for. Combine that with the slow response time of the TAs, and you have the recipe for some pretty frustrated students. Now on to the greatness of this course. This course truly felt like a graduate level course. There was a good mix of derivations and applications in the homeworks and the exams. There were 5 homeworks, each ranging between 3 and 4 questions. You got about 1.5 weeks to complete them, but you could get them done in a couple of days if you sat down and focused. If you could make your way through the homeworks, then the exams themselves wouldn’t pose that great of a challenge either. That’s not to say they weren’t challenging. The truly great thing about this course is the topics themselves. There’s a lot of great techniques and analytical methods presented in this course that should get WAY MORE of the limelight than some methods pushed in everyday machine learning courses. We covered things like functional data analysis, smoothing splines, b-splines, image compression, image augmentation, CP tensor decomposition, Tucker tensor decomposition, wavelets, matrix completion, anomaly detection via sparsity constraints, nuclear norm optimization, compressive sensing, signal decomposition, etc. Since I took the course over the summer, so we didn’t get a chance to cover the two optimization methods modules, but the lectures and homeworks were made available for self learning. But for what was covered, not only did we “cover” them, but we deep dove into the math as to why they are what they are, and why the do what they do. An anecdotal example is I’ve always known that L1 regularization drove features to have sparse coefficients in optimization. This course will not only show you why visually, but you’ll derive why mathematically too, and it’ll make absolute sense. Then you’ll also see other useful methods for L1 and L2 regularization outside of trying to fit a regression models. I guess I should add this part to the negative side, but it might be a positive for some, but be prepared to derive the closed form solution to multiple variations of an Ordinary Least Squares (OLS) problem. I could go without seeing a hat or a projection matrix for quite some time. Overall, do yourself a favor and take this course. Just take it by itself so you can truly appreciate it.

5 — ISYE 6644: Simulation

Difficulty:3/5

Enjoyment: 1/5

Time Commitment: 5 hours / week

This was my absolute least favorite class so far. They need to change the name of this class from “Simulation” to “Statistics and Pseudo Random Number Generators for Those Who Want to Contemplate the Theory and Pitfalls of Creating Their Own Simulation From Scratch (But Should Never Actually Create One Because There Already Exists Packages That Generate Simulations Better Than You Ever Could) With Extra Emphasis on Inverse Transform Theorem”. I’ll start off with the pros of this class, which is pretty much the same that over else raves about on OMSCentral. Dr. Goldsman is pretty entertaining in the lectures. If it wasn’t for his personality, I wouldn’t have made it through at all. He adds just enough comedy to make the lectures bearable. He really has a grudge against Justin Bieber and the University of Georgia however. The activity from TAs on Piazza was great and nothing was ever left to simple ambiguity whenever people had questions. And finally, everything was self contained. This was one of the few classes that you didn’t have to supplement the lectures with outside learning and resources. The course had 13, weekly homeworks that were multiple choice. They weren’t very tough, all you had to do was pay attention to the lectures. I generally knocked them out in a 2–5 hour sitting. Now for the negatives. There was little to no simulation in this course. There was about 1 to 1.5 weeks where you used the Arena simulation package. But other than that, NADA. Lectures consisted of some of the most deep level stats questions I’ve come across, and we are not talking about rolling dice, picking cards out of a deck, or running hypothesis testing. You were given crazy multivariate CDF and PDFs in which you had to double integrate to calculate marginal and conditional probabilities. There were Cholesky Matrices, brownian motion, thinning algorithms, Box-Muller & Cauchy distributions, hand calculating differential equations using Euler’s Method, Hand calculating derivatives using Newton’s methods, Tausworth & Linear Congruential Estimator & Desert Island Generators, etc. etc. etc. Let me give you a taste of one of the exam question and answers:

Just imagine midterms with 32 questions of nothing but those, and you get the gist of what this class was. Overall there was 3 midterms. They were somewhat cumulative. Each midterm primarily focused on the lectures covered right up to it, with a few questions from midterms past sprinkled in. However, you are allowed a cheat sheet. There was also a project which you could team up or fly solo. I did a blackjack simulator from scratch and built a Dash web application around it. Overall I just didn’t care for this class. I think it mostly had to do with my hopes for the class going into it. With a name like “Simulation”, I figured we would be doing a lot of that. Once my bubble was burst early on, I just never got motivation back for this course. And because of that, I suffered my first non A grade. I got a B =( . So if you have 0 stats background, 0 calculus background, you’re definitely going to want to avoid this course. If you’re someone who just loves stats, this is the course for you, because you won’t be doing any real simulation.

5 — ISYE 6740 — Computational Data Analytics

Difficulty:3/5

Enjoyment: 3/5

Time Commitment: 15 hours / week

This course is a required course for anyone doing the Computational track in the OMSA program. It is OMSA’s equivalent of the OMSCS machine learning course. It had some great aspects to it, but it also had some meh aspects to it. First, there were 6 homeworks overall. The first two were kind of brutal and took a lot of time. I think a significant portion of students dropped the course within the first 3–4 weeks because of these HWs. But once you got past these first 2, it was pretty much smooth sailing from there. The first one had a few parts to it, but the most intensive one was we had to code our own implementation of the K-medoids algorithm from scratch for image compression. If you’re familiar with K-means, then it wouldn’t take you too long to pick up K-medoids, but it definitely had some tricky components to it when coding from scratch. It is a very computationally expensive algorithm. I think I spent a week alone just trying to optimize my implementation to run faster. But I did enjoy it. Below is the output running my K-medoids image compression algorithm on an image with k = 2, 32, & 256 clusters

You can see my full implementation here. The second tough HW had us recreating the results on a paper utilizing an algorithm called isomap. It’s a method we utilized on 64x64 images of faces and were able to cluster them in 4096 dimensional space based off of facial lighting and direction.

But like I said earlier, once you got past these, it was pretty easy. With the 6 homeworks, you got two weeks to complete them. Every HW had 3 questions, although for 3 or 4 of the HWs, the 3rd question was a bonus question that you didn’t have to do. In total, the course covered: K-means & K-medoids clustering, Spectral clustering, PCA and Nonlinear Dimensionality Reduction, Density Estimation, GMM & EM algorithms, Optimization, Naive Bayes, SVM, Neural Networks, Anomaly detection, Boosting & AdaBoost, and Random Forest. There were no midterms, but there was a group project. Overall, I appreciate this course for how deep it went on some of the algorithms with respect to the mathematical rigor. And for almost all algorithms, we had to code them from scratch. Very few questions were we allowed to use pre made libraries (sci-kit learn). Because of this I definitely walked away with a much stronger understanding of these algorithms. My only complaint is some of these algorithms are just over done. Things like SVM, Random Forest, Logistic Regression, etc are touched on in so many other classes. There’s just a lot of overlap between this course and other courses. I do recommend this course, but if you struggle with mathematical derivations and closed form solutions, you might want to wait to take this one after you experience a few other courses. Also, if you have 0 experience in ML, this is a good intro course (somewhat), but you will see concepts from this course in others. So it won’t hurt you if you wait to take it.

6 — MGT 8803: Business Fundamentals for Analytics

Difficulty:3/5

Enjoyment: 2.5/5

Time Commitment: 2 hours / week

I was sitting in a bar in Atlanta early in the morning, the day of graduation, with an old friend I was catching up with. Somehow we got on the topic of starting a business and the amount of effort that goes into running one. But not just from a day to day stand point, but the bigger picture as well as the little granular details. I had him in thought provoking epiphany after thought provoking epiphany as I discussed things such as centralized vs distributed supply chains, marketing and business strategy, accrual vs cash basis accounting practices and how easy it is to inflate/deflate your company’s worth from simple asset estimates. And that was the pinnacle of this course (not the actual course itself). This course is the equivalent of an undergrad business degree crammed into one semester. Don’t be fooled by the title just because it has “Analytics” in it. You will barely be doing any. Unfortunately this is one of the required courses regardless of your track decision. I 100% emphatically, scream from the top of my lungs, beg the administration to change that. This should be an elective because I feel like I wasted time and money for something I probably won’t use, nor did I particularly care to pay $1000 to learn. I’ll start off with the positives of the course. The material was good. I’m not saying that the execution of the course is poorly done. There is a good mixture of pre-recorded lectures and weekly class meetings that were about 1.5 to 2 hours long. (If you can’t make the meeting time, don’t fret, they record them. You obviously just won’t get the opportunity to ask any clarifying questions during the session). The course is broken up into 5 modules, each one being 3 weeks long:

• Financial Accounting — This covered things like Assets, Liabilities, Owner’s Equity, Ratios, and Financial Statement Analysis. For the exams, we actually had to hand calculating/create the balance sheet for various companies. In order to do this, you had to know the difference between various assets, contra assets, liabilites, contra liabilities, and long term vs short term for all of the above.
• Finance — This covered financial management and investment rules, evaluating investment opportunities and stock valuation, calculating risk, expected return, cost of capital, and firm valuation.
• Marketing — This covered Marketing strategy, opportunity analysis, segmentation, positioning, buying behaviors, product development, place development, pricing, and promotion
• Supply Chain Management — This covered long, medium, and short term decisions, distribution strategies, forward and backwards vertical integration, etc.
• Business Strategy — This covered Environmental Analysis, strategy formulation, pestel analysis, internal analysis, differentiation, cost and focus leadership, etc.

As for the negatives, I just didn’t care for this course because I didn’t sign up to do an MBA, I signed up for Analytics, Machine Learning, and Data Science. That is my largest gripe with this course and it needs to be changed to optional. This only took up the space of all the other courses I WANTED to take but couldn’t because this had to be fulfilled.

6 — ISYE 6748: Analytics Practicum

Difficulty:N/A

Enjoyment: 5/5

Time Commitment: N/A

This is the culmination of all your hard work. This is where you get the chance to put theory to action. This is the applied practicum. In order to qualify for this course, you have to have completed at least 8 courses, and 3 of the 8 must be the foundational courses. Once you qualify, you receive an email from the department letting you know. You then must decide if you’re going to do an employer sponsored project or have a project assigned to you from one of Georgia Tech’s partner companies. Since I already work as a Data Scientist, I opted for an employer sponsored project. There are videos you have to watch as a part of this course, but for the most part, you will spend your time working on your project. There’s not much to say about this course except it is whatever you make it out to be. The work that I did for my project ultimately got published in a major journal in my industry. I even got to present at a Norwegian conference. The groundbreaking results I got, stemmed from actual methods I learned in my most favorite course ISYE 8803: High Dimensional Data Analytics. You can see the abstract to my paper here:

IMoDD: Intelligent Mapping of Downhole Dynamics

Conclusion

This program was worth every cent. Since being a part of this experience, I have not only grown exponentially in my field, but I’ve been published, I’ve been a presenter at Nvidia’s GTC conference, I’ve become a mentor for a bootcamp. I’ve done so much more than I ever could have fathom the day I first submitted my application.