ABATTBQ11 said:

---

Update: I'm a full year in at UTSA in their Data Analytics program, and so far it's been well worth it. I've learned a ton, and it's given me a lot of new perspective on business problems. I've also made a lot of good friends well outside my normal sphere. I'd definitely recommend the program.

---

if you don't mind, can you elaborate a little more on the new perspectives?

jpd301 said:

---

ABATTBQ11 said:

---

Update: I'm a full year in at UTSA in their Data Analytics program, and so far it's been well worth it. I've learned a ton, and it's given me a lot of new perspective on business problems. I've also made a lot of good friends well outside my normal sphere. I'd definitely recommend the program.

---

if you don't mind, can you elaborate a little more on the new perspectives?

---

It's more a mindset than anything. When you're creating models and assessing their predictions or the relationships between data that they describe, there's a lot of nuance and understanding that has to be there, and you have to look at it from all angles. You have to question whether your data is possibly biased, whether your model is biased, what kind of relationships might be built into your dataset and how they might affect your model type, how often different classes or factors occur in your population and what your base rates are, what tradeoffs you're willing to make, etc. I've seen many business decisions get made with fallacious reasoning because some of these things aren't taken into consideration or there's just too much focus on the desired outcome as opposed to the possible or probable outcomes.

Another semester down and another one to go (still holding on to that 4.0), and much of this rings true.

I use Power BI at work because it is easy to manage and quick to deploy, but it's not doing a lot of what SAS, R, or Python are in terms of statistical analysis and ML. If all you're doing is Power BI and Tableau, you're missing all of the good stuff.

So far, my biggest complaint would be a lack of building analytical tools into applications, but that is probably something to be worked on with a developer amongst a team. Also, big data applications like hadoop and distributed system concepts were introduced and explored from a theoretical standpoint, but we didn't practically work with them. I have one class (and an outside practicum left), so that may change though.

I wouldn't put too much value into Hadoop - it's dying.

Good news on PowerBI - MSFT is starting to build in some ML algorithms. Not sure when we will see them in regular deployment, but it's in the works.

That's what I've heard, but it's what is out there now. Really what I'd like more of is just the hands on experience. There's been quite a lot of R, Python, SAS, and a little GRETL, but nothing really with distributed systems. For what I'm doing, that's fine, but I might one day want to change industries.

I'm a little leary of MSFT's AutoML. It's automating sampling, normalization, feature extraction, model and hyperparameter selection, and cross validation. I'm not sure I'm comfortable with that.


ETA That kind of ML also puts you on a more equal footing with competitors. You can still choose what data to feed the models, but the decision making process is still the same. By pushing the decision making off onto MSFT's algorithms, you give up a certain level of control and ability to differentiate your internal processes. Not the end of the world, but for food thought.

ABATTBQ11 said:

---

Another semester down and another one to go (still holding on to that 4.0), and much of this rings true.

I use Power BI at work because it is easy to manage and quick to deploy, but it's not doing a lot of what SAS, R, or Python are in terms of statistical analysis and ML. If all you're doing is Power BI and Tableau, you're missing all of the good stuff.

So far, my biggest complaint would be a lack of building analytical tools into applications, but that is probably something to be worked on with a developer amongst a team. Also, big data applications like hadoop and distributed system concepts were introduced and explored from a theoretical standpoint, but we didn't practically work with them. I have one class (and an outside practicum left), so that may change though.

---

I'm not a data analyst, but I am in a professional MBA program at T20. I guess I'm the sucker.

Anyways, I use R and Python within Power BI for modeling. Although in Big 4, I consult in my old industry of construction and engineering. I pull a lot of data from P6, JDE and the cloud. Power BI pipes well to everything. Its handily to manage, analyze and visualize in the same platform.

RangerRick9211 said:

---

ABATTBQ11 said:

---

Another semester down and another one to go (still holding on to that 4.0), and much of this rings true.

I use Power BI at work because it is easy to manage and quick to deploy, but it's not doing a lot of what SAS, R, or Python are in terms of statistical analysis and ML. If all you're doing is Power BI and Tableau, you're missing all of the good stuff.

So far, my biggest complaint would be a lack of building analytical tools into applications, but that is probably something to be worked on with a developer amongst a team. Also, big data applications like hadoop and distributed system concepts were introduced and explored from a theoretical standpoint, but we didn't practically work with them. I have one class (and an outside practicum left), so that may change though.

---

I'm not a data analyst, but I am in a professional MBA program at T20. I guess I'm the sucker.

Anyways, I use R and Python within Power BI for modeling. Although in Big 4, I consult in my old industry of construction and engineering. I pull a lot of data from P6, JDE and the cloud. Power BI pipes well to everything. Its handily to manage, analyze and visualize in the same platform.

---

Not a sucker at all. They're both good routes. T20 MBA might be even better because it's a good route to the executive level. I considered an MBA and probably could have gotten into a T20 program, but at 30 with a wife and 2 (now 3) year old I wasn't moving anywhere to do it. My understanding was that MBA's had become a dime a dozen, so you really needed a prestigious one to stand out. One of the greatest benefits was the networking and getting to know lots of other up and coming or higher level professionals and increase your mobility. Not really something I could do online or remote.

UTSA's MSDA program was just a good fit, both personally and personally.

I'm actually in construction, so I'm very curious what kind of consulting you do and where. I don't incorporate R or Python into PBI because no one here would look at it. Anything I do analytically involves a lot of explaining and glazed eyes, so I give people what they want to see in their PBI reports and write the occasional white paper with some nice looking graphs, charts, and tables for anything complex.

Also, you say you're in Big 4. Any interest in a controller position?

What kind of knowledge and experience is needed to shift into data science / machine learning without a degree in the field? It seems there is a lot of demand for the skills.

bmks270 said:

---

What kind of knowledge and experience is needed to shift into data science / machine learning without a degree in the field? It seems there is a lot of demand for the skills.

---

There are plenty of people who call themselves "data scientists" because they have read blogs and technical documentation to set up models, but they're like kids playing with fire because they don't understand the meaning of what they're looking at or the mechanics behind it. You definitely need an understanding of stats, databases, coding, and model mechanics, off the top of my head. Database and coding experience helps with getting data and setting up models. Stats and model mechanics is important in model selection, feature selection, and model interpretation. I think a degree is worth it if you want to move into it because data science and ML are far more subjective than most people appreciate, and there is a LOT of nuance that the uninitiated don't understand or even realize exists. Without knowing the fundamentals and going through some kind of structured learning, it's very easy to make mistakes because you don't know what you don't know. There's also a lot of tools out there that are making ML more available and doing a lot of this for you, but the problem, to me, is that you have to trust the developers and you don't know what's going on under the hood. There are lots of things you could optimize for or make tradeoffs on, so how do you know that the decisions being made by the automated tool are the right ones for you?

The program I'm in basically required a STEM undergrad or an undergrad that had statistics in it. Social sciences degrees were also generally accepted because they had multiple statistics classes. Without that, you needed a few years of work experience in a field related to data management/analysis. I got into it with a modicum of related experience and a really good GRE score.

My first semester had a basic stats class that was very remedial for some of the more technical or experienced in the program (but a good refresher or crash course for everyone else). It covered different statistical tests, assumptions of those tests, and some fundamentals of modeling. Subsequent semesters went much deeper into how different models worked, their tradeoffs, how data selection affected models, how to assess models, the tradeoffs of different assessment strategies, how to set up and tune different models, and then practice on doing all of the above.

I already have an MS in engineering, not looking to go back through a whole degree program. I have python modeling and data post processing experience, executing test programs and drawing conclusions from data. It's hardware instrumentation data which I know is a different than the statistical data used for data science and ML. Behavioral science is also a hobby study of mine. Just seems like a field suited to my strengths, working with data and that a lot if it centers around behavioral modeling.

Is a "data science / machine learning" degree needed to get a foot in the door with an employee? Are transferable skills and open online courses enough to get in? I was thinking I could take a few open courses then pick a personal project or two (honestly I'd consider a paid university course or two but probably not a whole degree program).

you'd just need to look at job postings and see. More often than not, they aren't looking for a certain degree, but are more interested in actual job experience and software knowledge.

But, one of the keys is being able to interpret a model in the context of a business setting.

So can you design a data set? Run a multiple regression? Do you know if that's the right way to model the data? Interpret what that regression tells you from a business perspective? Can you tell if your regression model is valid?

bmks270 said:

---

I already have an MS in engineering, not looking to go back through a whole degree program. I have python modeling and data post processing experience, executing test programs and drawing conclusions from data. It's hardware instrumentation data which I know is a different than the statistical data used for data science and ML. Behavioral science is also a hobby study of mine. Just seems like a field suited to my strengths, working with data and that a lot if it centers around behavioral modeling.

Is a "data science / machine learning" degree needed to get a foot in the door with an employee? Are transferable skills and open online courses enough to get in? I was thinking I could take a few open courses then pick a personal project or two (honestly I'd consider a paid university course or two but probably not a whole degree program).

---

Depends on where you apply and what you're applying for. Definitely suited to your strengths it sounds like. You may not necessarily need a degree, but it would be a lot easier with one. You could look into a certification to show you know what you're doing and get your for in the door.

As I said and Vernada touched on, the biggest problem is that models and ML algorithms will generally give you at least some kind of result, even if you feed them poop flavored data. The trick is knowing how to look at your data to create features or remove data and how to assess your model to know if it's good, useful, or too good to be true. This is the part that you'd most need to know and become familiar with.

If you want a good place to start, get a copy of Elements of Statistical Learning (I might be able to send you a pdf copy if you want) or An Introduction to Statistical Learning. It's not exactly a hard read, but it is still a text book. IIRC correctly it will cover linear and logistic regression, LASSO, ridge regressions, Principle Component Analysis, Support Vector Machines, Tree based models and random forest, General Additive Models, and some basic clustering, as well as model assessment and the bias/variance trade-off. If you already have an MS in engineering, you should be able to understand most of what is in it pretty easily. It also has some good R walkthroughs and practice problems for the different models covered.

If you want to get into deep learning, try Deep Learning with Keras. It's a python based book that uses the keras package. I've had some assignments out of it, and it's pretty good.

So I am in a newly created "business intelligence" role in my company, and I've basically been given a blank canvas to do something with it. Where do you think I should start to get "smart" on this stuff? Should I start hammering through Python courses? Maybe some other EdX or other type analytics courses? Any thoughts would be appreciated!

A bunch of Kindle books on data science and programming today:

https://slickdeals.net/share/android_app/t/13753757

I've been in a similar position - my approach was to find specific projects ("actionable intelligence") to deliver. Then find the best tools to complete those projects. The tools have since varied (python, SQL, PowerBI, Excel, etc.) depending on the project.

In my opinion, a workable knowledge of a language (probably python or R), plus familiarity of at least one data visualization tool (PowerBi, Tableau, Qlik, Alteryx) is a good foundation for the role.

Thanks Sam!

What Sam said. I kind of carved out that role in my company and have done several BI "projects" so far. Somewhat simple stuff in BI terms, but big for us. I started out kind of learning as I went, then decided to go the masters route to get a better handle on the really big stuff. I think it's been worth it, especially with some of the professors I've had and their different experiences.

What I would do is identify needs across your company and develop a roadmap of what needs to get done and for who. Then look at where the necessary data will come from to get an idea of what you will need to get it from production to visualization. Then find a visualization tool to bring everything to. What tools you decide on will be determined by your overall goals and strategy. Different tools have different capabilities and price points, as well different deployment strategies.

If you're looking to do deeper analytics, you need to make the same decisions on what tools to use. SAS is very expensive, but well supported and secured. Python and R are cheap and open source, but they're also open source. There's also no telling exactly what is going on in packages you may use, so you have to have a certain among of trust in developers and the community. Power BI is beginning to offer analytics as well, as mentioned earlier, but it's more of a premium thing and it is making a lot of critical decisions that may or may not agree with. You still also need to do your due diligence on what you're feeding it and how, so you still need a high level understanding of what it is doing and what could throw it off/spoil your results.

Thanks ABATTBQ! All very helpful. Did your program teach you any programming languages?

I'm in the middle of getting my MS in Data Science from Northwestern. It's a lot of work but I've learned a lot. It's centered around Python and R with specializations for Machine Learning, Data Engineering, etc.

I had been teaching myself Python for about 8 months prior to starting and using it for personal/work projects. Mainly web scraping, API work, database interactions, with some machine learning. The combo of self taught and my coursework has led to my salary going up over 50% in the last year. I'd say these courses are worth it if you can pay and have tuition reimbursement. My company pays 5500 a year for that.

I'd say Georgia Tech is probably best bang for your buck but Northwestern has some great courses and specializations.

stonana said:

---

Thanks ABATTBQ! All very helpful. Did your program teach you any programming languages?

---

Sorry I haven't come back to this in awhile. Yes, it did. Gave you a lot of the basics, but up to you to find documentation and learn some of the modeling and more advanced functions and packages. Profs were available and not going to hang you out to dry.

Oddly enough, one of classmates (very smart guy), is a GT undergrad in petrophysics.

Final grades are in today for my program. Finishing with a 4.0. Not walking for graduation kind of sucks, but it's good to be done.

Overall, it was a good program and I'd recommend it. It will get better with time, and if you're starting now you'll probably have a better experience than me. My biggest complaint was too much work being done in SAS. This last semester, EVERYTHING was SAS based, and it was more of a course on how to use enterprise guide than Tools and Techniques, which is what it was supposed to be about. It turned into busy work on how to import data and transform variables, but in ways particular you SAS. Theory on tools and techniques was limited to left, right, inner, and outer SQL joins. Even then, you created queries in SAS's ****ty UI instead of learning the actual SQL. Query optimization would have been much more appreciated and useful.

That stinks. When I did my program there was an EG course, and although you have to spend a lot of time learning the program, it was coupled with text mining/analysis so at least we were exposed to those theories and concepts.

But, originally, my program was very heavy SAS. That has changed since my cohort and they use a more broad array of programs now.

Oh and CONGRATULATIONS!!!

Outstanding achievement!

I look back and still can't believe I CHOSE to put myself through that stress and misery (of course I'm glad I did).

For that sweet spot between SAS and pure coding of R/Python, check out Alteryx.

Significantly cheaper than SAS. Easy enough for basic automation tasks for the Excel heavy user but powerful enough for an advanced, true analytics user. Plus you can plug in R and Python code if ever wanted/ needed.

Alteryx.com

We did text mining and analysis in python. You have to develop the automations yourself, but it's free... Doing it that way also forced you to understand the underlying theory.

Honestly I didn't like EG, and few people who used SAS even used it. They all did code, though I found SAS code very foreign. Our program director was not thrilled when he learned that we'd been doing as much EG as we had, and then had us do a market segmentation in SAS code using an old Experian data set. Not exactly my favorite assignment.

Thanks! My wife deserves a lot of credit for keeping the home front and being patient. It was very challenging with our opposing schedules and a toddler, but she helped out a ton

Congrats on the accomplishment. Been following this thread for a while. I decided to enroll in a small Data Science course with Columbia University's Emeritus program this September. Great starter course for Python, the statistical portion was mostly focused on code. Had to spend time to read through the theory and understand all the different distribution types. We are going to get into ML soon. Course was $2300 but is making me heavily consider getting a masters in data science. Fun stuff!

Is Max Kilger one of your professors? I had him in my MBA Classes

Congrats. I graduated in 2019. Although not with a 4.0

I use PC SAS primarily for my work and love it. I don't generally use SQL language in it however as I like thinking within the scope of the process data vector much better than SQL's filter functioning. I also think EG and SAS Studio are terrible so if you were trained on that, I can understand why you don't like it.