Path to a free self-taught education in Computer Science
github.com
github.com
One criticism: I believe that the majority of programming jobs are software engineering jobs (even if that's not in the job title). Since many are using CS as prep for programming careers, I believe this list may need more emphasis on software engineering and contemporary skills and knowledge in SE.
For example, I saw testing in the advanced section if I remember correctly. But it was one of multiple options I think. Doesn't everyone need some testing knowledge/experience?
Another one is requirements analysis or negotiation. Figuring out _exactly_ what needs to be done, in a way that makes complete sense all the way through, is feasible and high priority, isn't nonsense from a UI or database perspective, and getting the other people to go along with that, is pretty important to most programming jobs I have had.
Another example: back in the good ole' days of CS, large scale code re-use may have been mostly a dream for most popular programming language ecosystems (that weren't Perl). But now, even though dependency reduction still seems much cooler, being able to use real package management systems and select the right modules or components is critical for most jobs. Not sure that type of thing gets quite enough attention in this list.
It seems to me that the basics of OOP, the kind of things that most programmers need to know about, can be covered in a single lecture or two as part of a general programming or software engineering class. Java-oriented design patterns and general API design seem more appropriate as an advanced/optional offering for students who are interested.
The linked curriculum has 6 weeks dedicated to Object-Oriented Design in the intermediate part of the "core programming" section. It seems like that much time could be spent on something more fundamental.
I’m a CS professor. My colleagues and I have complete control of the program. When you see an outdated program it’s the faculty’s fault in general and the leadership’s fault in particular for not forcing the issue.
1. (Intermediate/Core) I'd cover basic OO in Python. I'd demonstrate basic features like instantiation, introspection, properties and methods before moving on. I would not cover defining new classes before I had covered creating a dispatcher with first-class functions in a hash-map/dictionary. I would not cover defining new classes unless I could cover pros and cons of using OO style versus simple functions that accept rich data structures like lists and hash-maps.
2. (Intermediate/Optional) I'd offer Java, C# and(or) Smalltalk classes at the intermediate level, where the goal would be competence with the chosen language's implementation, tooling, ecosystem and common programming idioms. These classes would involve object-oriented principles, given how important the paradigm is to these languages, but it wouldn't be the sole focus of the class. Edit: Also C++ if possible.
3. (Intermediate/Optional) A design patterns practicum. (Visitor, observer, factory, etc.). Just a programming-heavy class where you solve problems using as many of the common design patterns as possible.
4. (Advanced/Optional) A bottom-up class where we'd cover implementation details of object systems in a low-level systems language like C. Perhaps a semester-long project might be to build a working object system in C. The pedagogical goal would be impart an understanding of the kinds of problems that object-oriented programming was invented to solve by relying on first-hand experience rather than dogmatic instruction and hand-waving about the evils of procedural style.
5. (Advanced/Optional) An Object Oriented Design class that would cover issues in designing large-scale systems using object-oriented programming. It would cover issues relating to inheritance, Object-Relational Mapping, and API design. This is where you'd cover UML and that sort of thing.
Someone would need to start this Sophomore year and not miss a single course until graduation in order to complete this... While that's fine for core classes like Chemistry which have multiple timeslots offered per semester, having such a brittle pathway for a less-taught topic would cause some grief to plan around.
I would argue that a large codebase at any company is going to be impenetrable to most fresh computer science graduates. Production software architecture almost always has idiosyncracies that they can't teach you in college, no matter what paradigm is used. So you are always going to have to learn on the job in those situations. You'll pick up the necessary object-oriented principles then.
Trying to teach OO programming in to newbies in a classroom environment usually leads to a lot of incidental complexity and dogma that feels like tedious make-work to students because it is. Meanwhile, with appropriate guidance and peer review from a senior programmer or software architect, and an opportunity to observe and participate a real production build process, an OO design is likely to just "click" in a relatively short amount of time.
One of the key points of OO is that it is a discipline. OO involves explicitly coupling data to logic in encapsulated chunks with a clear separation between the API and implementation details. Trying to teach that particular discipline to students who still can't accomplish anything meaningful with code is, I believe, a waste of their time. OO features like public and private methods seem obvious when you actually have an API to maintain, whereas for a student learning the fundamental techniques and technologies it seems like an arbitrary and largely pointless distinction.
Encouraging and/or enforcing coding discipline is the responsibility of senior members of a given team and is most effective when practiced that way.
Most software today is written in an object oriented language, and for large software systems the quality of your object oriented design has a huge impact on how much it costs to maintain and improve over time. I've been writing software for 20 years and I've never had to implement hash table or search algorithm, but almost every day I'm thinking about my software design using object oriented principles. Personally I think a 6-week course is too short.
We could debate the importance of teaching search algorithms, too. But it's a different issue. Search algorithms are a solved problem[1]. Algorithms like that, in addition to being mental exercises, programming practice, and preparation for a career in academic research, are usually used to demonstrate complexity analysis and space/time tradeoffs. The space/time tradeoff is a problem you will face in nearly every environment no matter what language or technology you use.
[1] As far as we know... there might still be research on searching algorithms but for the purposes of the vast majority of software engineers, sorting functions aren't something they'll ever need to implement.
However I don’t think knowing the gorey details of transistors is necessary. What they need to be aware of is cache hierarchies and pipelined superscalar processor architecture.
These sophisticated architectural features can be flouted by code that causes them to behave poorly. Such as code that does random accesses in a large buffer causing the probability of the desired data being stored in cache to be low.
The ways in which real computers differ from idealized ones is what CS students need to know.
Also CS students ought to have to learn to read others’ code.
I did go and get an online Masters of sorts (an MBA from WGU), which was definitely do-able while working full-time (and also useless), but it was nowhere near as rigorous and time-consuming as an engineering graduate degree would be :)
You work with PhDs, so can pursue your graduate interests directly from them while being involved in the research, and get paid at the same time albeit not industry standards, prepare for lower salary. Open AI hires developers too and plenty of researchers are there.
Having seen what poverty can do to someone I have always put securing my future financially over my education, in the hope that one day I can return to university and resume.
And I am one of the lucky ones who could afford / was explained benefits of an education from a young age.
Most of the value of an MBA is in the network, not the knowledge so distance learning and correspondence degrees are worth drastically less. You can get an MBA from the UK’s Open University or Queen Mary’s, University of London for a little over twenty thousand pounds or thirteen thousand tespectively.
It sucks he's a great guy and one of the hardest working people I know. He also blames his english. I've heard a lot of people with English as a second language use that as an excuse and, like him, it's usually from people who tend to speak and write English better than a lot of native English speakers I've met.
Since then I've offered to show him some of the machine work I do and have tried to explain more about programming the machines to him. He seems excited to learn so I think he'll enjoy stuff he can learn at his own pace.
The funny thing is I've actually learned a lot of things about computer science written by Iranians or other people who English was not their first language. It's a barrier but not as much as i've heard so many people make it out to be especially if you know enough to read, write and communicate.
You can learn about the idioms such as Callbacks and Hooks, but to really grasp how these are organized into frameworks seems difficult for authors. It's easier to just tell you, If you want this effect, Do that. Or, I'm just not getting something about computer science pedagogy; Maybe the domain of programming where I find my interests doesn't teach programming the way I think about learning it. I will dig into this to see what's here.
Software engineering might cover design patterns, but even then it's a limited set of techniques with broad application - idioms in practice are tuned to their language and execution environment, because clumsy idioms don't propagate.
The only way I've learned that was to just start making things. Whenever I ran into a problem i'd look up whatever I was having trouble with. It failed a lot. I've had to go back an restart things more than once or endes up giving up on things that were dead ends. Then after a couple years of that I stopped having to quit because i'd completely messed up the idea of what I was doing.
I honestly have no idea if the things I've learned are correct or not but programs I write seem to do what they're supposed to now.
I think the reason why there's not much about how to put programs together is that it's pretty much up to you. There's a million ways to do one thing and most programs have a million things that need to do one thing. You need a good idea of what you would like these things to be. Then you pick one and do it, then another one and another one making sure the first things you made still work with the new ones. If you don't know how to do one of the things either look up and see how other people did it or move on for a bit and come back to it.
A framework, library or program is just input data, either from you or a user, processing of data and output. It's up to you to figure out how to structure these things for the needs of your program.
There is no right way, just whatever way you find that works. You can look at a program like building a machine. The programming language you learned gives you the nuts, bolts and basic materials you need, the algorithms, callbacks and hooks are the different prefabbed parts and you're the mechanic that bolts it all together with a wrench and a whole lotta grease.
For a trivial analogy, consider this template for a C program:
#include <stdio.h>
#include <stdlib.h>
int main(int argc, char *argv[])
{
// Your code here.
return 0;
}
Frameworks essentially do the same thing on a larger scale, and usually with multiple discrete components. The key difference between a library and a framework is that a library is something that the developer imports and invokes, whereas the framework establishes a useful context for a specialized type of development.
The vast majority of us never use more than basic algebra. Counting, multiplying and dividing money, and estimating effort in story points (and maybe dollars as you get seniority) are the big math items.
I'm horrible at calculus, a fault that cost me around $30,000 in makeup classes and a delayed graduation as I kept losing scholarships.
Yet I have never, and will never, use Calculus in my field. This was told to me by the career coach in my intro class, who said almost none of you will ever use Calculus, yet my school still made everyone take up to Calculus 4. Who the hell is finding the area under three-dimensional curves without relying on a library? Madness.
A.I, machine learning and distributed systems will become more in demand to replace it.
The bulk of the work is still connecting the algorithms to real world data sources, creating interfaces to monitor and view their results, and explaining to non-technical people how to use them.
If anything AI/ML will create more jobs for non-math people who are working on the glue level because the better the tech is behind the scenes the more value it creates and the more potential customers there will be.
Me? I regret that I don't know nearly enough statistics.
Beyond that, I wish that instead of Calc 2 and 3 they would have had more heavy programming courses.
On the other hands, I think you don't need computer science to be a programmer, and a lot of us have suffered and will suffer from the coflation of these two fields. Many "computer science" curricula are actually programming. Many people who want to learn programming enroll into computer science because the difference is not clear at all. This was the case for me, and I was lucky to actually like CS, but it was not what I thought I had signed up for.
I don't think the parent said zero calc or statistics (rereading the parent did say no calc which I disagree with). I think the complaint stems from the fact that many CS curriculums started, and often still exist in the math department. Should a CS student take calc classes up to level 4 or stop at level 2 and instead have more advanced courses around the computer part of the science?
This is a resource to learn Computer Science - not computer programming. Calculus and many other advanced mathematics topics are very much needed if wishes to learn Computer Science.
Also, for some reason, I found abstract algebra related to objet oriented design - similar kind of thinking.
This is free course - they won't be in unfortunate 3000$ lock you was. (I think calculus is useful, but I don't think it is worth paying that much money.)
Yes! And that's why you need to learn to be fluent in: Spanish, Chinese, Russian, French, and some other languages. You never know when you need them!
At the time I was in school there was a real worry of outsourcing all programming, so I didn't want to be 'just a programmer.' I hoped in the worst case, if all the jobs were outsourced I could be in charge of a remote team. Turns out the outsourcing didn't happen quite to that level (although at one point in the past I was in charge of a teams on the other side of the world), but the business courses have been very valuable over my entire career so far.
I'm seeing this type of split more often in CS, and I think it's great. CS + business or CS + art or CS + media, CS + whatever I think is the future. Programming alone is great (someone has to write the libraries we all use ;) ), but take a competent programmer, add another skill they are passionate about and you have a real force.
I’ve actually recently started exploring Linear Algebra on my own. So seeing this list is helpful, as I’d like to explore Calculus too.
I agree with your point, though. My courses in Economics and Accounting have proven valuable to my career so far.
You'd certainly go crazy if you start to think in all the little details as soon as you face a problem, and probably never deliver a solution on the time needed.
It turns out that mathematics are very good at, gradually, giving you new levels of abstraction to solve new problems, or old problems in new ways, etc. The important is not to memorize formulas (that's what books and wikipedia are for), but to know which one to use, and when, their uses, etc.
I love derivations and proofs, and majored in math, but they are a deterrent to most people, and unrelated to their jobs like you say. I work for a company whose products are particularly mathematical, yet only a tiny handful of programmers (mostly the ones with science backgrounds) deal with the math related stuff.
Most engineers finish college, start their first jobs, and immediately become so busy with CAD and bureaucracy, that they forget all of their math and theory.
I think another problem is: What to do with kids who want to become programmers, and have been told that they must get a college degree in something, but programming doesn't really require 4 years of college study. Many of those kids major in Computer Science, which many people have pointed out, is not the same as programming. But selective admission into Computer Science programs becomes self fulfilling in terms of the demand to get into those programs. I took a different route, which was to major in math and physics. And I ended up doing something other than programming for my career, but that's OK too.
I assume by "order of magnitude" you're talking about big-O and friends... I use that math all the time when I deal with things at scale. If you're doing anything that's not a CRUD app with a trivial amount of users, you quickly run into math.
By the way, I am wondering if you had some bad teachers who themselves failed to fully understand calculus. Good teachers make all the difference.
I've been amazed at how many programmers I've interacted with in the past two decades who cannot operate outside the one, small subject they currently know. It's one thing to say you can't work on scientific modeling because you couldn't do Calc, but when I see front end developers who say "nope I can't do back end ever" I realize that they were never challenged and they will be the last person I want on my team. I want to have an employee who says "I don't know that topics but I'm going to learn it and solve this problem!" Dumbing down our profession doesn't help us.
I used Calculus one day to solve a random business problem.
I noticed the pattern was a Taylor Series.
It had been over 10 years since I learned it, but I immediately recognized the pattern.
I solved it via brute force on Excel. It took over 50 pages of Excel calculations to reach the answer. (About 2500 lines of calculations, which is easy to do in Excel.)
And once I noticed the pattern, I was able to turn it into a formula.
And now, I had a one-liner formula, that I was able to use to solve my business problem. Then, I plugged this formula into Excel too, and now I had a little Excel app that could get the answer for me.
Did this make me any money? No. Did this directly make the business any money? No. But, it did help the business by now having a mathematical formula that they can use to solve that specific question, if another customer asks for it in the future, specifically during the initial sales process.
And all this, because I had been forced to study Calculus in college.
Beyond that, computer science is an offshoot of mathematics (Turing laboured on Computability, Boole on relation algebra, Von Neumann on finite difference methods). To rid computer science of mathematics makes about as much sense as stripping psychology out of counselling,
You have a much too ’practical’ view. What you have in mind is not a graduate degree in computer science but some kind of... vocational training for programmers. Which is OK, if that’s what you want, and provided you leave the faculty of computer science intact for those who wish to study the abstract field and make progress therein.
When it comes to sitting down and writing software, such a mathematical computer science education provides a clear benefit. The more math you know, the more problems you can formulate and solve with a computer. If the job is cranking out web pages, a mathematical background will increase efficiency because you'll recognize the problems at hand fit into familiar formalisms with familiar solutions. Moreover, if you run into a hard problem, a mathematical background will provide the necessary tools to work through it in a principled manner.
Really though, the immediate applicability is beside the point. We are not talking about learning to program, we are talking about learning computer science. There is value in having a breadth of understanding in basic math topics (yes, calculus is a basic topic; if you want something advanced you need to look at real analysis).
If you're writing code that deals with list-shaped user data, you better believe you need this, because how the code scales with n is a constraint on the business.