Can someone explain how you can train Data Scientists in 6 months? What jobs do these grads walk into?
The tech prerequisite is: "You should feel extremely comfortable with how a computer works including: touch-typing, web browsers, search engines, and basic computer programs."
The Curriculum:
Stats, Linear Algebra AND Data Wrangling: 4 Weeks
Modeling: 3 Weeks
Data Engineering, Databases, SQL, Productionisation: 3 Weeks
Machine Learning, NLP, Neural Networks: 3 Weeks
Python, OOP, Algorithms & General CS: 3 Weeks
Project work: 3 Weeks
At that pace over so many subject is anyone walking out able to remember what a dot product or a t-test is from week 1?
You assign precourse work that rejects 98% of applicants. That helps select students with existing experience, sometimes even STEM degrees.
According to the example contract on their site, their income share agreements last up to seven years. Even if they fail to teach a student anything, the student can spend years studying on their own (or even get a degree), and still owe money.
>You assign precourse work that rejects 98% of applicants.
I think a good model for this is the Insight Fellowship. By selecting mostly STEM Ph.D. graduates, the pool of students already has some relevant background/foundation. The focus can then be on the tools and methods for applying DS in a corporate/business setting, which is actually relevant for jobs.
Insight grads, in my experience, have been a notorious joke. Nothing has diminished my respect for a PhD from a top-tier institution more than the people I have worked with coming out of that program.
Interesting, I've had the opposite experience. One of the best managers I've ever reported to and worked with came from the project. I've also worked with a few good engineers and scientists from there, too.
I suppose this is a reminder of The Law of Small Numbers and sample size importance!
if you have a STEM degree, what value can you derive from short introductory classes on linear algebra, Python and relational databases?
Countless truly excellent and free/cheap resources on these topics exist and you don't have to give a way a significant percentage of your yearly salary.
I think the value is similar to the value of a personal trainer. There's no rocket science involved in maintaining fitness, but many people find value in the added accountability and planning support.
serious question: Why isn't there a service "personal trainer for coding"? Not so much for juniors looking to get into the industry, but... I'm sure there's a continuous pipeline of high-value people who would pay $$$ to keep them accountable to doing say three coding exercises a week while they are getting ready to switch jobs or are between jobs.
Perhaps because one of the "three virtues of great programmers" is hubris - which Larry Wall optimistically defined as "The quality that makes you write (and maintain) programs that other people won't want to say bad things about." but which also/really means "Excessive pride or self-confidence."
I suspect many/most coders think a "personal trainer for coding" is a great idea - for other people (especially their idiot co workers). But they obviously don't need one themselves, and would never need to pay for new...
Hubris drives you to re-build things you shouldn't... which is how you learn why they work the way they do. And maybe to try things you "shouldn't", which end up working out OK fairly often.
This actually exists. I would link to a specific example, but it's run by an individual and I'd feel weird doing that. "Software coaching" might be a useful keyword?
There are programming tutors on wyzant and other services. I suspect a lot of people learning to code who already have a coding friend/person in their life do this informally.
I really like the idea of a talented tutor helping with subjects I struggle with. Even if it costs thousands of dollars this can be a great investment. The offer would have to be focused, customized and beyond the basics. Given the materials that are out there (MIT open courseware comes to mind) it is really beyond me why people spend so much.
You can procrastinate away your entire life with "someday I'll surely take advantages of all the free resources on X that are out there and finally learn X and maybe even shift careers!"
Just think of all the things you never bother to learn despite having a latent interest even though you can open a new Youtube.com tab right now and fire away. It's hard to start, and then it's overwhelming when you do, and then you feel like a failure when you don't stick with it.
Isn't it worth quite a lot to snap out of it and finally do it? It doesn't surprise me that people find this to be worth thousands of dollars. When you consider how fatally expensive it is to procrastinate your goals forever, maybe these schools are a bargain.
This is literally me, although today I finally found the motivation to actually put together two of the parts I 3d printed a while ago. Maybe some day I'll finish the robot.
I've thought about the Lambda School model, sure the funding model is good, however the problem is what that funding is being used for. If you could work on the material on your own and hire a TA paid by the hour to grade your assignments, give feedback and provide assistance, basically only pay for the services that you actually need then the cost for a degree would go down massively.
What Lambda School is trying to do right now is just fish for good students and hope they graduate as quickly as possible so they can collect as many ISAs as possible for as little cost as possible.
Exactly. Some people need a structured environment to get that kind of work done at a decent pace. It also helps to have feedback from a teacher, just like a personal trainer helps you with form.
I have seen another approach (also with STEM PhDs), where it isn't an income share, but rather that the incubator gets a referral fee if the student is hired--similar to a recruiter. Now, why do this? The math behind most of data science is rather basic for a number of PhD physicists--I would say the greatest weakness on the math front is that many will know probability, but not Bayesian statistics. However, the level of coding will be mixed with many not knowing about source control and a number coding in say Matlab rather than Python. Now, some have said that people can learn this on their own, which comes to the next reason. Physics graduate students and postdocs can put in extreme hours and their supervisors can expect those hours. 100+ hour weeks happen (especially for experimentalists--a number enter industry and it takes readjustment to figure out what to do with spare time). By formally leaving and joining an incubator, they give themselves space/time to learn--it's also a pretty intentional act where they have to decide that they are leaving the field (which can be psychologically difficult). The next is job interviews--the typical physicist has no idea what a data science interview will be like and coaching will help them a lot. Finally, there is networking, where the incubator may have connections with hiring managers where they can at least get people into interviews. I have seen people do either the incubator route or the DIY route and for those who have gone with the incubator (where they don't pay and the business model isn't a percentage of their salary coming from their pocket, but rather a recruiting fee going to the incubator from the company they get a job at), then they seem fairly satisfied with their experience.
Exactly this. The only Bootcamp grass I know that were successful in their post grad jobs were already developers but with a different skill set. The entire concept of turning non technical people into developers is a farce.
> Can someone explain how you can train Data Scientists in 6 months?
Doesn't it ring alarm bells that society attempts to herd 100% of students through 4 year programs? That sounds far from optimal, no?
It seems highly likely to me that if you get someone who excels on the steep parts of learning curves, 6 months studying the right things would be just as effective (if not more) than 4 years doing mostly busy work.
> What jobs do these grads walk into?
I think they walk into entry level jobs. I know I much prefer working with people that love to live on steep learning curve slopes over those who prefer to go at the pace of the system.
> remember what a dot product or a t-test is from week 1?
I have been in data science for many years and most of the terms I was taught college (like t test) never come up. Maybe once or twice a year I have to spend 60 seconds to refresh on dot product algo.
>It seems highly likely to me that if you get someone who excels on the steep parts of learning curves, 6 months studying the right things would be just as effective (if not more) than 4 years doing mostly busy work.
I don't know what you were doing in college but anyone who can learn four years of material in six months probably doesn't need Lambda School in the first place.
If you cut out the commute, obsession with socializing, 1x speed inefficient lectures, impersonal general purpose homework, breadth first class catalog that try to get you well rounded - the necessary material for most professionals is not four years worth. "I never use what we learned at school, professional life is very different" is a very old meme.
Where I live, colleges prepare you to be able to choose to be many things - continue academia, or be part of the workforce under diverging titles. They assume you don't know what you want to be yet - and it is mostly true. But if you are at a point in life where you know what you want to be, you can skip most of it (or even do it live when needed). The "keep your options open" attitude of colleges have a lot of overhead. What most people need from a 4 year college can be easily condensed into a year. It WON'T be 4 years worth of material. But if you know what you want (not keeping your options open) you don't need most of it.
I mean, I’ll be the last person to say you need four years of college to be an effective programmer, but that’s not really what I take issue with. I dropped out of my BS after freshman year and got my first job in the industry instead at 20. The rest of school isn’t busy work, I just prefer teaching myself, as I had been already for eleven years.
I saw countless people at my cow college who were utterly unqualified to be there, but because "It's free money!" and Mommy and Daddy expect them to go, they went.
For entirely too many of the young women, it was Four Year Husband Selection Tryouts, and for too many of the young men, it was Random Sexual Encounter Weekend. A couple of folks were there because we had the most powerful supercomputer in American academia, or for our excellent agricultural and engineering programs, but they were the exception rather than the norm.
Going back at age 28 after a six year stint in the Navy gave me a very different perspective though, so I may not be the best example.
Statistics of enrollment percentage doesn’t really describe the culture surrounding college in the United States, though. Plenty of people don’t end up going to college but that certainly hasn’t stopped it being considered a panacea for employability in this country.
This is still being pushed by politicians. It’s still what pretty much every parent hopes for their children as well. On this website I’ve seen people derisively describe people as not college educated.
> On this website I’ve seen people derisively describe people as not college educated.
This is unfortunate, because I've met quite a few very smart people who never went to college, or who started but had life circumstances that pulled them away before they were able to complete a degree.
College is not a panacea and the sad but honest truth is that most jobs do not require a college degree. Most of the jobs in America can easily be done with on-the-job training; we've just seen companies shift to a model where they want to offload those costs to the American worker by letting them spend $50,000 to $250,000 and four years of their lives "proving" they can get an education that many of them don't really need, and don't really want in the first place.
We also allowed well-meaning people to pass laws preventing companies from conducting IQ tests and thereby sorting people as was done many years ago. I know this is true because years ago I used to eat lunch with the Director of Human Talent Acquisition at my company almost every day, and we would discuss this. He was a very engaging gentleman in his mid-70s with an almost encyclopaedic knowledge of current and past HR topics, philosophies, and controversies. He had worked for a well-known Fortune 50 company for almost 35 years before "retiring" to my company, a mere Fortune 1000 player. He can recall using IQ tests to sort applicants into positions for which we now demand bachelor's degrees, or in some cases, even master's degrees.
A bit sad, if you ask me. Our culture should be progressing, not regressing...
It should but it is likely that the alarm will not be raised enough. We are ALL taught from grade 1 to "prepare for college, so it can turn us into adults" as a life philosophy.
It probably does not help that every single one of our teachers was forced through that system in order to be granted the "privilege" of teaching us.
The signal to noise ratio is so awful these days that it is a serious liability to understand what you're talking about in data science interviews.
I have seen been on hiring side of multiple larger organizational interviews where the "answers" to the questions asked during interviews were not even correct. Candidates that knew what they were talking about would be rejected because clueless interviewer didn't know what they were talking about.
Whenever I have to interview I have to play guessing games with interviewers trying to understand what they think is the correct answer. I've had to explain that, yes you can do statistical inference with a linear model, that Poisson regression and Logistic regression are in fact only different in terms of a link function, and of course also had to explain why a Poisson random variable might help in modeling number of purchases. In all these cases the correct answer was "use XGBoost".
I think most people who would be considering bootcamps are comparing it to a traditional 4-year college curriculum -- at least I did. I enrolled in the full-time web program a few years back and this was the back-of-the-napkin math I used to make my decision:
Total Hours of College:
4 years, of which 2 years are "core" coursework and 2 years are general education.
2 Years * 2 semesters per year * 15 weeks per semester * 15 credit hours per week = 900 hours for a degree.
Lambda School:
6 months * 4 weeks per month * 40 hours per week = 960.
Obviously this doesn't account for a number of variables such as 0-credit "mandatory" labs, out-of-class work/projects/studying, but a Lambda student who spends an extra 5-10 hours per week during the program would end up with 1000-1200 total hours.
But this still ignores the main advantage of a program like Lambda, which is the ISA. An education that would be inaccessible to students at the margin -- think people who can't afford to move to a university, qualify for loans, or take 2+ years off of work -- can get in if they're committed enough. Even if Lambda were a total scam, at worst you're out 6 months and $0. If Local State University doesn't get you a job, you're out 4 years and $50k+.
Looks like you made a calculation error or misunderstood your units. A credit hour is 3 hours of work, as a credit hour is "(1) One hour of classroom or direct faculty instruction and a minimum of two hours of out of class student work each week for approximately fifteen weeks for one semester or trimester hour of credit, or ten to twelve weeks for one quarter hour of credit, or the equivalent amount of work over a different amount of time"
So it's 2,700 hours for a 4-year degree, versus 960 hours in Lambda School, just using your method of calculation. It's also not counting the internships or summer programs that students in 4-year schools usually partake in. And it does not count extracurriculars during the schoolyear, like hackathons, interview preparation, programming competitions, student group projects, etc. Finally, you're assuming that an entire half of a college degree is geneds, which is really not the case. It's more like 1/4 geneds, 1/2 required major/concentration courses, 1/4 electives which many students opt to take technical courses in. So probably more like 3,200 (minimum) to 5,000 hours in a 4-year college.
General education courses do not take up half the course work. Moreover, you are suppose to spend 2-3 hours working out of class for every hour in class. You realize you have to study and do homework outside of class... No one considers a bootcamp to be comparable to an degree at a good university.
I have trouble getting students to concentrate on difficult material for anywhere near a whole class even if it's their only class of the day (and I'm considered a good and entertaining professor, honest). No WAY I could get anyone to concentrate on difficult material for 40 hours per week. I mean really no way, even if I were a teaching genius.
Learning difficult material means taking breaks and digesting stuff subconsciously.
Therefore, learning anything difficult takes a certain minimum of calendar time as well as hours in the classroom.
Enough buzzword to pass an interview. Unless someone already has a pretty solid degree, it's impossible to learn the fundamentals that quickly. And if they do, they should take grad classes on data science, not go into a bootcamp...
They have. There are companies in the UK that hire Oxbridge grads from the humanities with no tech experience whatsoever and train them. In the States, IBM has a similar apprenticeship programme that lasts for 1 year.
In the mainframe era (e.g. up until the late 1990s maybe?) the big US consulting firms did the same thing. Andersen Consulting would hire smart history and english majors and teach them COBOL in 6 weeks and then dispatch them to a client site to work 60 hours/week. Granted their approaches, designs, and code were highly standardized, but the work got done.
I did this. I started in QA with very little experience. I had some linux experience from personal projects I'd worked on and 3 months of a pretty mediocre python developer bootcamp. The company I started at mostly hired less experienced people and had them work QA for a year to get them integrated. It worked really well. You were on an integrated team with QA/Dev/Ops and as QA you sat with other team members and did manual testing while you figured it out. They had a great mentorship program and were serious about training and promoting from within.
I think one of the reasons bootcamps get away with this kind of thing is that, for at least many people, traditional schooling also fails to teach this material. So if you're going to just teach people enough to get a job, but not actually to do it, better to do it for "cheap" in 6 months than, say, expensively in a 2 year masters program.
I think the problem above is probably somewhat worse for data science than software engineering in general, but it applies broadly to quantitative jobs in the tech industry, in my experience.
Data science bootcamps don't make much sense to me unless the student is coming from a STEM background. If someone has studied engineering and already had courses in linear algebra,stats, basic programming, then it shouldn't be hard to learn the more practical data science stuff
I kind of suspect it's a bit of a light-weight Pareto principle: you can gain 70-80% high-level knowledge (note that I said knowledge, not understanding), which is enough to get started with some real-world applications with sufficient handholding and mentoring.
I don't think this is necessarily a bad thing, but expectations should be set appropriately.
If you've pre-screened to select for candidates who would be able to take that and run with it, I'm not suprised that you could achieve a sufficient amount of decent outcomes.
I think this is correct, re: the Pareto principle.
I'm a Lambda grad (Full stack web, not DS) and at the end of the program, I was knowledgeable about a lot more than I understood; and that also includes being knowledgeable about a lot of my shortcomings that I was painfully unaware of before Lambda.
A year after Lambda working as a SWE and I am now finally starting to feel a sense of domain... I hesitate to call it expertise, so lets say domain 'comfort' when it comes to building and maintaining react or react-like FE applications and node BE apis. I've had a mentor at my company since I started and mentorship is necessary for anyone coming out of a bootcamp with no prior industry experience, IMO.
The expectations set within the bootcamp are reasonable, IMO. Instructors never made it seem like we were going to fully internalize everything we were being bombarded with. Practicing what you learn is crucial to turning any of the things taught into a skillset, and that is made very clear.
I imagine the outcomes are pretty good for those who get through the program, as well. They were great for me and most of my cohort peers that I keep in contact with. The ISA model, IMO, makes a lot of sense and I'd be surprised if Lambda couldn't make it work in the long run.
A friend of mine got a data scientist job at a well-known tech company after a summer-long bootcamp. He had previous experience, though, doing some academic-track work in a scientific field. So the training was giving him a more solid background in an area that he was kind of informally nudging into in the first place.
Technically the prerequisites are very loose but the best success stories I hear are usually more like, someone got a STEM degree but actually not all of those degrees translate into well-paying jobs, after this boot camp they were able to shift over to a data science career.
That seems like an odd course setup for a DS. I'm not a DS but kind of learn about it as a hobby, but I think if you cut the course down to just the first 2 topics people would have a better foundation.
All the DB related stuff seems like a bonus, ML should be easy to pick up if you have good LinAlg and stats knowledge and Python you can pick up along the way.
I also agree on your last point. I don't understand how someone would be able to remember a fraction of this knowledge when you're changing topics so often.
The last point is important. When you're in a university you take different courses over the years but they often reuse or are based on some of the material covered in earlier courses. You learn by repetition and as you go into more advanced topics you can often understand better what you learned before - because you learn connections between things.
But when you need to learn a lot of things in a short period of time you don't have time to reflect and think how it all works together.
I find this often in my programming projects. Each sub-project requires a lot of focus and concentration. That seems to make it easy to forget about the previous (sub-)project almost totally. Putting in lots of new stuff into your brain means some older stuff gets pushed out of there.
The more you learn the more you forget. The only way to fight this is to try to understand the connections between the sub-topics, but if you are in a hurry you don't have time for that.
I don't know data science topics, but I can tell you a friend of mine was working on a Computer Science BS while I was in Lambda Schools CS section. He had a semester each on Algorithms and Data Structures. I had 1 month on those 2 + Graphs. At the end of my month, I was teaching him these things because his Data Structures class was basically SQL.
I can't directly answer your question, but I'm sure the answer has something to do with figuring out and teaching the base requirements of the topic, methodically, with well experienced industry professionals.
> Stats, Linear Algebra AND Data Wrangling: 4 Weeks
Ridiculous. Introductory statistics and linear algebra are two separate undergraduate level courses. And if you want a real grasp of statistics, you need to learn probability theory first, which is another semester long course.
To say nothing of "data wrangling", whatever the fuck it means.
Your intro stats and Linear Algebra classes are 4 hours a week. These classes are 40 hours a week. Every week is essentially a quarters worth of work. You can take undergrad courses like that in the summer.
Pretty sure "data wrangling" boils down to god-tier regex knowledge - which is less theoretical heavy-lifting than an acquired art that comes with practice.
It is more than that and it is what Data Scientists spend a significant amount of time doing. Real world data is extremely messy and a lot of time is often spent understanding what exactly the data is, changing it from wide to long, deciding how to deal with missing values etc.
This highly cited paper is a good introduction to the subject:
There’s decent science on why this isn’t possible. Basically you need to space things out to give the brain time to work out what to put into long term memory
It's possible, but for very few people. I went to MIT and there were stories of people there taking ~12-16 classes a semester which theoretically averages out to a course a week (though they ran concurrently and it's a crazy course load for the average person). To be fair, the people who could do that were way on the right of the distribution and it's hard to convey just how fast they could pick things up (think of them as the equivalent of an NBA player vs the average MIT student as a D1 athlete).
I react well to that approach if I'm really excited about the material, and usually do well in the coursework. It does take a few months for everything to "really" sink in though, and having a lot of friends or acquaintances from MIT and Stanford that did that I think it's rather the norm.
Approving coursework doesn't mean that the concepts sunk in. It just means you have new tools to further your understanding of everything, but it still takes a long time for it to sink in. During that time maybe you just need to do nothing, just let it sink in, but in my experience everyone ends up needing some time.
You simply can't take 12-16 classes a semester, the day doesn't have enough hours for that. They _might_ have gotten permission from the professors to skip classes and just show up to the exams, which is _not_ learning.
Not sure why that's not learning - if they understand the material at the time of the exams, who cares if they show up to class? Most people forget most of the stuff they "learned" in college 10 years out anyway - most of the economic worth is in the credentialing / signaling versus the actual education (since getting into the school is the hardest part)
I meant that it is not learning in the context of the discussion, which was whether people can assimilate that amount of information in a semester. If you show up to the exam and pass then you did all the learning prior to that.
Depends. Mathematics? Chemistry? There's some room for compression there, but not much. If you ditched all of the absolutely useless "general education"[1] classes, and kept class sizes small enough to allow time for individual coaching, I think most STEM undergraduate degrees could fit nicely within two to three years.
[1] Students are welcome to "expand their horizons" when they aren't paying thousands of dollars for the privilege of being forced to do so.
As a Data Science student that just recently graduated from the program, it prepares you for a data analyst position, at best.... which essentially means any job where your primary role is to wrangle data.
One issue is that if you take a top CS program grad with “proper” academic training and even some internships, they still need to look up references or take a quick review to remembers the things in this list. What really made them competent at work is the ability to learn quickly and solve problems not in book.
That is the main concern I have regarding these bootcamps is that they focus way too much on surface knowledge.
There a wide range of data scientist roles. I’m a senior data scientist and couldn’t tell you how to do a dot product or a t test
A lot of my role is just analyzing data for patterns , implementing rules into our product, building models using very basic libraries and putting them into production, monitoring models in production.
It isn't as though 4 year degrees teach purely useful information. If you stripped my computer engineering degree to courses that I use for software engineering, it would fit in two semesters. If you tossed out all the non interview required theory, it could fit in one.
It is a bit like playing tennis. Playing against your 75 years old limping neighbor is easy, playing against competent people is hard, playing on the tour is for the very talented only.
The tech prerequisite is: "You should feel extremely comfortable with how a computer works including: touch-typing, web browsers, search engines, and basic computer programs."
The Curriculum:
Stats, Linear Algebra AND Data Wrangling: 4 Weeks
Modeling: 3 Weeks
Data Engineering, Databases, SQL, Productionisation: 3 Weeks
Machine Learning, NLP, Neural Networks: 3 Weeks
Python, OOP, Algorithms & General CS: 3 Weeks
Project work: 3 Weeks
At that pace over so many subject is anyone walking out able to remember what a dot product or a t-test is from week 1?