Recommendations For Noob In Data Degree

Hi!

I am changing fields from a social sciences background into a more stem-driven computer science / data sci one. In an odd twist, my advisors are not from the comp-sci department and cannot advise me on issues I have.

I have been told the future jobs related to my degree will be from Data Analytics / Science fields, and I will be expected to know Python and R 'well' before graduation.

I have taken basic calculus for social sciences (non-simplification Calc 1 basically), but not Calc 1/2/3. I have taken social science stats courses (econometrics, public admin calc), some basic data stats courses (data mining, Analysis using R), but feel wildly unprepared. I effectively have no Comp Sci background.

Most of my peers come from a Comp Sci (pure) or Economics background.

• I had the very minimum requires to get into the program. But I mean, like the equivalency of plain Ody Gear without any augments 'minimum requirements'. Sure I am 119, but that's like it and ppl within the degree question why I am there lol.

Albeit, I currently have a 4.0 from spending TONS of time tying to learn the material.

In an attempt to NOT be useless mathematically, or via coding, I wish to audit courses to make up the gap in my knowledge.

Does anyone have any recommendations for courses I could, or should, take? Any recommendations in general? :)

Thank you,
Senaki

Open source LLMs will be better at writing Python code than any human by the time you finish training to a decent level.

R is used in statistics mostly, and as soon as people accomplish making LLMs good at Python and other more common languages, they will train them in R too.

I don't think data analysis or programming are very AI safe career paths at all.

Data Science is a wide field that sits on top of a stack of technologies that were developing individually overtime.

But overall, its quite easy to grasp the full picture. For that, I would advise you to enroll some course about CRISP-DM, thats also called data mining, but its the basic overview of the entire pipeline.

In that pipeline you will see the more specific fields: IT infrastructure happens heavily in the process of ELT and Big Data.

Data architecture and business intelligence are major part of the corporative subjects related to data.

Math, statistics and programming are the base of machine learning.

Machine learning is the basis of AI.

The current challenge in most companies is integrating AI in the business intelligence to provide quality insights for decisions.

But AI is heavily based on data quality, thats related to the data architecture ELT Big Data processes.

In a nutshell, thats data science, the art of making meaningful use of the zillions data that companies produce.

The job prospects are better than my first degree...(which is basically abject poverty).

Is it worth while to take courses pertaining to Big Data (ex: learning Hadoop), or Data Infrastructure management (ex: SQL courses).

I've noticed a lot of courses seem to focus on ML Algorithms of some degree. Ex: Random Forest, Regression Analysis, KNN, etc. But do not seem to require the math behind the occurrences (at least with regression). Not certain why they focus so much on this.

Those courses are less popular than ML so they tend to have higher demand than the course where everyone is bandwagoning (IMO). But I dont think particularly useful to learn some of those technologies unless thats exactly what you will work with or if thats something you actually want to work with, so you can prepare for an interview and say thats your strong point.

Like I said, ML is a stack build over those math statistics and pregramming.

Being build over those, you dont actually need them to perform relatively well in ML because most of the heavy lifting was already done in those ML libraries, where you just need to write a relatively simple script and have the results.

But going one step further would require you to understand whats behind scenes. But thats like going one step further in direction to a PhD, for example, not for everyday work (ML is a refinement of statistical techniques, model solving is a computation processing using calculus and derivatives).

As a last note, current ML is almost like an art than actual science. Thats because despite having written library to get results, most of its (hyper)parameters comes more from heuristic than a method.

Supposely knowing more about math and statistic can help you create better models, that are more representative of your domain, or that can be solved with less effort/faster. Not knowing them will make your tuning more experimental than theoretical (IMO).

Sadly I'm still too fresh into the degree. I am still not even sure what 'end goal' I want as of yet, so I cannot predict if I will use it or not. :(

Makes sense. I'm going for a MS.


I see


Thank you for all the info. I appreciate it! :)

You don't need R, just Python.

The joke used to be that data scientists are people working in industry as statisticians without statistics degrees. Now the joke is that actually no one cares about statistics and data scientists are just Python devs that write bad code.

I have a BS in comp sci, with a minor in mathematics and a speciality in computer architecture and design, which is basically an intro to computer engineering. I graduated in 2019. These are some of my experiences going through it.


I actually don't think this is an odd of a twist as you think. My advisors while going for my comp sci degree were not in the technology field at all and are mostly there for the bookkeeping aspects of your schedule. My advice here is figure out the chain of command and talk to the lowest rung that has the degree you are after. In my case, I was in the office of the associate chair of the comp-sci department, and one of my regular teachers was the associate dean of technology, and I could go pick their brains about stuff. This sorta depends on how big the school is and all that, but sorting out my specialty woes was a lot easier w/ the chair of the comp sci department instead of the advisor w/ the communications degree.

Edit: Also, I had advanced knowledge that the school was planning a math minor and saw a couple iterations that took, because I knew both the chair and associate chair of the math department.


Are calc 2/3 required for your degree? In my case they were shunted off to the minor, but calc 1 was and it was the great filter. Biggest piece of advice I can give here is they don't really all build off each other in order, 2/3/diff sorta all flow from calc 1, so if you do only ok in calc 2, like I did, you might still be good in 3.


If you learn this way great, but there was no bigger waste of my time than slamming my head into the book. What I ended up doing to reinforce the material was tutoring the classes I had already done. I intially did this with the gen ed classes after impressing the chemistry professor.


This can bite you in the ***. If you're doing it because "job good" you're gonna hate it, and as others have said it's a big field so you can miander into a bunch of different topics. The specialty I ended the degree with isn't the one I started with (which was biometrics), but after getting an advisor to voodoo my schedule to include the assembly langauge class I liked it so much I went back to school to push even closer to computer engineering. The two classes where I picked up a soldiering iron were more fun than half the classes for my degree. It doesn't have to be a super concrete goal, but you are gonna want at least a cardinal direction of where you're going.

When I was deciding my degree I was talking to the chair of the information security and intelligence degree at Ferris here and he has a minor in biology he never uses.

You might wanna pump the breaks here. I had an ex-girlfriend that almost had 3 degrees but switched at the last minute each time, including 1 degree at an expensive Christian school. Combined with above you might go through various options and decide to scrap the whole thing.

I'm not gonna be much help here since you seem to be focused more on data science topics, which I would have merely touched on. Python and R weren't a part of my degree at all though I did learn Python for my internship. My general degree used mostly C#, and my specialty used assembly and C++. The few biometrics classes I took used matlab.

I know a lot of this is Debbie Downer stuff, but I watched people flounder here, get filtered by calc 1 or logic, and I myself changed specialties once. And I'm not much help on the job side of things either, sorry.

Edit: All of this had Veterans Affairs breathing down my neck the whole time, so any sort of schedule changes had to be approved simultaneously by the school and the VA, so you probably have more wiggle room.

Being able to work with people and understand society seem like safer future bets than working with data in the age of AI. I'm pretty sure the "learn to code to get a safe career" train has already left the station.

My post was mostly facetious. A good number of my cohort ended up working entry-level non-BA requiring degree jobs tho. To which they cannot afford to live well now. So I wouldn’t recommend my initial route to anyone spare you speak Mandarin or ‘know someone’.

I have been loving the courses thus far and find them fascinating.

You’d be surprised by the lack of jobs based around what society ‘needs’ vs. why society ‘wants’.

Almost all of my ‘successful’ friends from undergrad changed fields except a select few who went to law school or teaching, or for a PHD. I guess a few spoke ‘in-demand’ languages and went for a high paying government jobs. But the vast majority really did not end up well off. I didn’t want to teach grade school, but was considering PHD when I found because I enjoy research.

Ended up picking this one because I loved stats in undergrad and figured I’d enjoy the field. Thus far I was correct. The level of math is a tad higher than my current tier, and coding is entirely new to me. But I’m getting there. :p

The public LLMs cannot code anywhere near as well as non-coders think they can. They are essentially predictive text algorithms, so if the problem you are trying to solve already exists in a near identical format on a dataset, they'll spit out a great solution. But, since most things need to be integrated with or even built around other existing non-public code, the model will not be able to solve the problems correctly. Further, it won't be able to tell you that it's not solving it correctly, so you may end up putting in code that appears to make sense but doesn't actually do what's intended.

I don't disagree that it's no longer a safe career path, though. LLM will likely be used by the best software developers to reduce the workload. There is a real possibility that in 10 or 20 years, it will be cheaper for large companies to train LLM on their own codebase and the languages and libraries they work with and use only a couple of human developers in conjunction with that. But, it's far from guaranteed. Outsourcing is an issue too, and if you're looking for a career that will guarantee 30 or more years of work.. I think you need to involve either social skills or physical labor. Almost everything relating to statistics, math, IT, and possibly even applied forms like engineering will become extremely competitive when AI takes the lowest rung out of the picture.

I was actually looking at getting a teaching certification as an alternative option while I have the free time and money to do additional coursework [dual purpose, my partner teaches so we'd have similar/same days off, but it's about as secure a field as it gets].

Back in 2016 my ex at the time was working as a translator for Booking, and had for almost a decade, in a language with about 2.5m speakers worldwide. Booking had about 20 translators since each and all offer on the website needed to exist in all languages. The translation toolkit they used was offering the english version on one side, and an empty box to write the translation on the other side

In early 2018 what Thorny describes is exactly what happened: Booking switched to an AI translation service, and the translators found that about 95% of the translations were already done, tho in a lot of cases the translations were poor (about 30% success rate). Translators became more "correctors" than anything else. Then in mid 2019 a huge jump in machine translation happened, and most texts were suddenly 97-99% correct. Fun fact: translators worked on commission per word translated, and didn't get any credit for any text box they didn't touch. Paychecks went from about 2-3k/mo to less than 1k almost overnight.

Then Booking downsized the office from about 20 translators to 5-6. I would be surprised exactly zero if the same thing happened at first to code monkeys - i.e. those developers hired to write menial code in large quantities - then shortly thereafter to mid level programmers. Those in charge of designing systems have maybe a bit more of a career, but we're moving to a world where humans are supervisors more than makers. And you never need as many supervisors as you needed makers.


Society? Job prospects have been driven by what corporations want/need for the past three decades at least. Society went out of the picture a long time ago.

Societal needs are filled by few overworked and underpaid individuals, who do it from a place of "***, this need be done or society crumbles", which is then exploited by the people who pay to be able to pay as little as possible. Financial data analysis is entirely unnecessary for society, and is done by individuals who aren't going to work for you unless you cover them in money.
Capitalism is broken, and has been for a long time.

My first degree offered job prospects almost exclusively within public employment.

I couldn’t agree more.

There is a reason why teachers are underpaid.

I have some experience at NGOs so I will use them as an example. NGOs, particularly the ‘humanitarian’ ones, typically try to focus on a goal to ‘do well by humanity’. Issue is, no one cares and no one with money will fund them. So they rely heavily on government aid. Often qualifying because it ‘looks good’ for politicians to throw some pennies at them. Their income as employees
Is about 20-30k a year.

How good is 30k a year considered?

Abysmal. In parts of the US, you can clear 40k/year doing service work like flipping burgers.

My previous work last salary was something around 50k a year, after exchange, and that was after 12 years working there, and its a very good salary by our standards.

My current job pays 40% less as a trainee in programming.

Im basically being paid less than flipping burgers.

Well, at least i work 3 days at home, so at least that

Would keep in mind that exchange rates don't tell the whole story. I would assume your general cost of living is lower there, so the actual buying power of your salary is probably a good bit higher.

[Of course, our borders are open, you could always come here and get paid!]

I think AI will do a lot to change the way development happens for higher level projects, but the lower you get (esp with embedded devices), the less likely it is to have a huge impact in the short/medium term IMO.

I've observed across my career that folks with experience with C#/Java/etc are pretty common (Python to an extent but surprisingly less) and easy to replace, similarly adapting AI generated code to these languages tends (but isn't always) to be easier. OTOH folks with expertise in anything lower level (read: asm, C, C++, and arguably Rust) tend to be harder to replace especially with the quantity of embedded devices growing by the day. There's a certain architectural knowledge that is difficult to replace.

I've worked in companies with thousands of developers and a small % of those knew even how to navigate the Linux kernel source, much fewer how to actually make meaningful changes without breaking something. These used to be core fundamental CS knowledge subjects that a lot of programs have neglected in favor of higher level languages/technologies, but those lower level skills/knowledge bases are essential for embedded devices in particular (well, everything, but it's a smaller pool for Windows/Apple/etc). A lot of the people who started 30 years ago when these were mainline CS subjects are aging out and leaving a void that makes it hard to hire people with this experience, background, and willingness to actually do the work. Most people give me some dead expression when I tell them 90% of my development is in C, like I just told them my cat died or something, but I've observed my peers having fewer issues with job retention and finding work.

It's not for everyone, but it's something to consider if working with embedded devices is something of interest to you.

Personally, though, if my kids asked me what to do: I'd tell them to go be a welder, blacksmith, carpenter, etc. People doing quality, dependable work in these areas linked up with good builders are doing very well.


Idk where you are but most academic programs require a terminal degree in the US, at least those I've looked at. Teaching certification programs you can probably get away with fine, but any academic institution is gonna look for a high level degree. Personally I'd love to do this also and I've taught many classes over the years, but doing it professionally is something I wrote off because I'm not going back to school again just so I can teach


My wife was an elementary school teacher 15 years ago in a low cost of living area and made more than this.

A lot depends on where you live, 100k here is pretty mid-level where I live, but in parts of the country it's a lot less, while in others it's a lot more. The cost of living, particularly real estate and food, makes a huge difference in how far your salary stretches. It also makes a huge difference in how you are perceived by banks when applying for loans or accounts.

My state has a 1 year program (with a BA as a prerequisite) that grants both a master's degree and the certification required to teach in the state. I would be looking to teach high school[math, probably], not find a competitive position at a university. So, when I said certification, it was a bit of an oversimplification.

They also pay teachers quite well as is and have a strong union guaranteeing raises that exceed inflation. I don't think it's the highest paying job I could get in a similar timeframe, and I likely wouldn't even consider it if not for my partner already teaching in the state. The benefits of both having summers off are pretty substantial compared to any other option, though.

Math underlies everything. And the difference between BA / BS & MA / MS is calculus and physics.

If you can a linked calculus / physics course take a year's worth. Such linked courses use the math you are learning right then and there for the physics and are a great help understanding and retaining both. Having lost the majority of my advanced math due to disuse I will say you WANT the maths and you want to retain them.

I think LLMs reveal an already-present problem with knowledge work, which is that it is needed the most by those who understand it the least.

IMO the real threat to US/EU data science in the next 20 years is just outsourcing to Asia, not LLMs.

This is the case at my company, a lot of our senior firmware engineers are retiring, and on their months long wind down they keep begging the company to hire their replacement now so they can train them but that person is either hired 2 week before the retirement date or never hired at all. More and More we are contracting out our firmware development and basically eliminating the internal path from entry level to master.

I think it is short sighted, especially in our specific field where not many other companies will foster the kind of experience we need in our technologies, we can hire consultants and contractors to work on our firmware today BECAUSE THEY ARE LITTERLY THE GUYS WHO JUST RETIRED! this is not a viable path forward.

Pretty much everything said in any of your classes is wrong. Do what you have to do to pass the class just understand that the professors have no clue how stuff works in any industry that isn't education. There is also an incredibly high likelihood that your career path will have absolutely nothing to do with your degree (IT related).

Get very comfortable self teaching, creating self projects that require you mastering a new set of skills. When mastering skills don't use shortcuts, don't cut corners, learn how to do high quality work that doesn't explode when placed under stress. This is what outside employers are really looking for.

This so damn much. Almost all key pounders are now either H1B's or remote workers straight from India. These guys work for a fraction of what a recent college grade thinks they are worth. India has become to IT what China is to factories.

People thought I was dumb when I said image generators would change my field (Design) in May 2022...

People said gen AI would never make videos, music, textured 3D models, and on and on, the nature of coding makes it highly vulnerable. Let's revive this thread in 12 months and see where things are rather than theorise.

Yeah and most of them are probably using AI to do the base code already. Over 50% of office workers in India already use ChatGPT daily.

Let me restate this for you "I am an expert in design and told people AI would change my field, I was right". "I am not an expert in programming/software development I told them AI would change their field, like it did mine, I was laughed at" Maybe if you had some expertise it would not be comparing apples and oranges.

The most I have gotten out of AI as part of my workflow is documentation, and it is amazing for it. I can do the manual work of documenting some flow diagrams using mermaid and set those as examples then pass in the whole code base and get out a solid flow diagram for each function and method. After that you can even ask it to generate flow diagrams for specific variables so it shows you how that variable is changed.

but outside of that actually doing the coding, it has done nothing more than absolute crap, which most code on the internet is in fact crap so no real surprise. It can not properly iterate or maintain context for a program that is in the thousands of lines of code. And it really is just polluted with too much crap to generate nice clean code, just look at how many damn comments the thing puts in.

Unlike the issue with image generation where it was messing up hands but everything else is pretty solid, there is a huge gap between the desired output and the current performance. Where art has a subjective aspect "eh close enough" where you can say the average person may just not care about the details that luxury does not exist in software.

Citation needed.

AI doesn't generate anything, it is neither artificial nor intelligent, just a really nice marketing buzzword. All it does is take input keys and the find the result with the highest probability based on those input keys, we've known how to do this for decades but didn't have the processing techniques to do it on a large scale. In the case of pictures, the AI dataset has several indexed versions of everything from tiger, to raid, to finger and so forth, this is what the training process creates. You make a statement, it deconstructs that statement to extract key words, then matches them inside the dataset and references those indexed images, stiches them together and applies the filters it things you want.

What we call generative AI is actually closer to a stochastic automation system. A junior level human could of done each of those steps, instead a machine running code does it. This can have amazing cost savings, but comes at the expense that you need a much senior period to review and fix the output. There is even a meme for it, a section that would take a dev ~2hrs to code followed by ~5hrs of debugging / fixing could be done with AI in 5min and take ~24hrs to debug / fix.


They try and I kick their crap back every day.

I am not exaggerating here, yesterday afternoon we have a meeting where we go over the tasks and components of the new system we're building. I'm the lead architect and responsible for putting everything together into a singular business system and therefor I essentially write the standards for how the various components integrate together, especially legacy cause everybody has legacy crud laying around. So we get to one of the dev for the transactional apps on the website and lo and behold he decided that the company is going to change default timeouts out the web access tier because sometimes his application would take longer then 10s to complete a request. This planned change, that he had zero authority to do, would break HA across the entire customer website.

Yeah that was a very uncomfortable conversation when I had to tell him in no uncertain terms would anyone be altering those values.

Everything CoPilot does can also be done with a simple google search and looking through StackExchange, Medium and some referenced github examples. Senior folks know this and have an entire library of already built components they can reference and copy paste edit into existence. Thing is, all those examples are largely for sandboxes or proof of concepts, not for use in production environments or when integrating with the bazillions of existing technologies.

AI isn't sending their best. I'm gonna build a firewall before it starts eating the dogs and cats