Berkeley offers its data science course online for free
news.berkeley.edu
news.berkeley.edu
We live in 2018 and there is no open source course for everything. Instead there are probably 10-30k universities who have similar courses and professors who give the same lecture every year.
They get paid often enough by countries to create and do those courses. In germany most of our unversities are paid by all of us germans anyway.
And what do you find online? Always the starter verion like 101 computer science or videos with bad audio or video, no proper exercises, no solution helper etc. Nothing. You have to go to different sites to sometimes pay or sometimes not.
there are no local locations to meet up with people.
There should be a global initiative for global free and open access learning. Sponsored and supported by companies and countries. Build upon a core of a knowledge graph based on topics or 'snippets of knowledge'. Like for example: math -> add -> sub
Something like 'The Map of Mathematics' (https://www.youtube.com/watch?v=OmJ-4B-mS-Y)
And when you wanna get the global accepted math 101 level, you have to take specific topics / snippets.
And those snippets can than be filled with different people who are making a lecture for that topic and you can choose whom you like more or who is better in explaining it to you.
What do i do instead? I ask around for the lecture scripts because they are always behind a simple password protected area or have multiple links to different pages of different universietses who offer different courses for free as videos for there students in sometimes/often bad quality and / or bad video players etc.
It sucks and this is stupid.
I like this idea and the framing of it a lot
I'm not looking for the next github page with collections of tons of different sites with different courses.
I still hope for one platform where all those smart people out there are working together to optimize learning.
The problem is that there are a lot of people getting paid a lot of money all over the world to work in post-secondary education whom control the keys to accreditation and whom have proven very resilient at resisting any optimization efforts.
Also need done good pr I guess :)
“Just add a bunch of green up arrows and red down arrows, your manager will love it” was advice from a co-worker of mine. Sadly, she was right.
It didn't actually give correct results, I'm not sure it could have worked (I needed to do two updates on the same cycle and never did figure out how), and I documented that it was buggy.
But it looked cool and I got good marks. So my experience pretty much agrees with yours.
After you have given a university hundreds of thousands of dollars and a decade-plus of your life, you will then be ready to teach the next crop of students.
1. You really think Berkeley's top-ranked PhD program is recruiting people who couldn't find jobs? No. Not only can 100% of successful top-tier PhD applicants find jobs, 100% of them are strong candidates for the top echelon on entry-level jobs.
If you disagree, go look up the people in Berkeley's CS PhD program. Point out a single person you think didn't turn down mid-100s job offers to attend Berkeley.
Getting into one of these top-5-to-10 PhD programs is no small thing...
2. MOOCs aren't PhD programs and in nearly all cases aren't designed to feed into PhD programs.
3. Finally, at least in CS, at least for the moment, educators are in extremely high demand. And again, at least for CS, that demand isn't being manufactured by the academy.
It is well known that PhD programmes churn out far more PhDs than can reasonably be employed in their field.
I mean, including friends and former colleagues I probably know maybe 300 people with PhDs. Of those I can count on my fingers those actually doing research in academia, probably one hand those on tenure track. But do you really think any of them slogged through the programme in Physics or Biology just to get a software job writing CRUD apps, or prettying up BI reports?
a) didn't bother applying but would've been shoe-ins, or else
b) knew very early (freshman year) they were research-bound and optimized for a non-industry objective function (but could've skated into an industry job of their choice given a shift in undergraduate career focus). E.g., couldn't pass a coding interview and no industry internships but have one or more top-tier publication in a hot subfield.
But (b) is kind of stupid to think about. It's like saying a successful lawyer would not make captain in the military. This may or may not be the case, but either way, who cares?
So (b) is slightly less trivial than saying a successful lawyer would not make captain in the military because most lawyers are one conversation, a few signatures, and one oath away from being a captain.
So don't delay lawyers, join today!
> ...So (b) is slightly less trivial
I'd argue not. In both cases, a fully capable and prepared person has to jump through a couple of relatively trivial hoops. And when those hoops aren't possible to jump through it's a weird case. We can split that hair, but it's silly.
They are mostly economics or (I’m not sure what it’s called in English, but it’s a degree in societal administration), but they really ought to be data scientists because everything they do is based on huge sql data sets.
We pay private contractors a lot of money to turn our data into cubes and manageable models because none of our analytics know how.
In 10 years I suspect anyone with that job title will need data science on their resume. Not just to manage the data, but also to start doing machine learning on it.
By comparison we have one network guy to run the network for 10.000 employees and 5000 students, with a backup guy who knows everything the first guy does but works with something else, you know, in case the first guy quits.
Do you propose that they should take a MOOC like this one, or that they should be replaced by MOOC graduates?
Intro to coding is a fascinating one. From a marketing/business standpoint it makes sense, but 3-4 it was extremely frustrating to see dozens of intro coding courses, but practically nothing for intermediate programmers. Thankfully we're past that point for the most part.
I guess it's analogous to the data science degrees popping up today. Will be interesting to see if it ends up as a fad degree or a legitimate career path.
I remember when Dolly was cloned and we would have a whole new industry...
What is the field that is going to revolutionize curing diseases through genetics engineering etc?
Companies are need a store front, so front end work will continue for a while, until it is super easy for any joe to make a professional website.
Data Science looks promising too, because it is automating and solving problems that previously could not be done.
However, outside of tech the world still needs skilled blue collar workers. For example, I don't see carpenters being automated away any time soon. I hear some of those jobs pay better than tech work too.
It is also a possibility that Data Science is an easier topic to learn than Computer Science, and thus more popular.
Seriously though, I think people are drawn to data science out of a desire to create stories with some underlying support (data/evidence) in order to influence policy or business decisions.
Internally, Berkeley definitely isn't based toward the intro-level stuff. Quite the opposite. But the most polished, rehearsed, mass-manufactured classes are certainly the gigantic intro-level ones everyone takes.
Is it that there are more courses in data science relative to other topics, or is just more marketing around these classes? It costs Berekely essentially nothing to pump out some press around the release of course materials in data science.
/Cognitive scientist
Deep learning is the current shiny toy but I suspect we'll find it isn't actually sufficient for a lot of things we want to do and we'll run into a wall a lot of people aren't expecting.
But I think I'd like some kind of formal, credentialed program that would build on my existing linear algebra + software skills (and address the weaknesses in my statistical understanding that I know are there based on how I felt about my grasp on the related material for even the classes I passed)... and maybe isn't quite as big an investment as a full-fledged master's degree.
Anybody have any suggestions?
It is a big commitment - 6 months full-time or one year part-time.
Depending on what stats you want to do, there are some pretty decent MOOCs. No one is going to claim that Daphne Koller's PGM course is weak in anyway for example[1].
[1] https://www.coursera.org/learn/probabilistic-graphical-model...
Which are the hard courses at Berkeley in CompSci using the site you linked?
If my memory holds, there’s a policy that class averages should be around 3.0-3.3 (B/B+).
I think for data science: Stats 134, CS 189, EE 127, EE 126 are most useful (in this order). Of course, in order to do well in CS 189 you need to have a good background in probability which can either be CS 70/Math 55 (if very well understood), or Stats 134, or EE 126.
This is bad in two ways:
1) The people taking these courses do not learn much for the effort and time they spend. Also it gives them illusion that they know enough as they take course from big university.
2) Industry is already so confused in hiring, they hire by name. So even you take these courses and study in depth on your own you can't get hired. Someone more qualified can not get hired just because they can't pay 100k to get a degree in machine learning from one of these big university.
This is really a bad trend and we should spend time on real courses. Everyone knows that TV series are waste of time, these courses are like TV series. Stop watching them.
Even if you believe it's pointless it's pretty clear it's no something everyone else "knows".
Seriously, just upload the lecture videos, put the homework online and textbook. Add a message board and you're golden.
Before Coursera, i was never able to finish anything on MIT opencourseware. Free flow of information need too much commitment from my end to be digerable.
It was the structure given by
> "registering" for the class and then following a regimented schedule.
that i managed to start and finish. Disclaimer: I discovered Coursera after grad school
I know the above probably sounds like sour grapes, I don't have an MBA or any graduate degree, I just think the whole Exec. MBA thing is a total scam against the companies paying for them and a huge cash cow for universities.
(There are two ways you can follow the course: Certificate Program is paid, but the AUDIT program is free)
We have a course (right a school application of stuff taught in school!) with two teachers, that is, two sections of the course, each section with its own teacher and its own students. At the end of the two courses, that is, the two sections, we want to compare the teachers. So we give the same test to all of the students from both courses.
Suppose one section had 20 students and the other one, 25 -- the point here is that we don't ask that the two numbers be equal; fine if they are equal, but we're not asking that they be.
So, there were 45 students. So, get a good random number generator and pick 20 students from the 45 and average their scores; also average the scores of the other 25; then take the difference of the two averages.
That was once. It was resampling. Now, do that 1000 times -- remember, we have a computer to do this for us. So, now we have 1000 differences. If you want, then, "live a little" and do that 2000 times. Or, for A students, do all the combinations of 45 students taken 20 at a time. Ah, heck, lets stick closer to being practical and stay with the 1000.
Now, presto, bingo, drum roll please, may I have the envelope with the actual difference in the actual averages of the actual scores in the two classes.
If that actual difference is out in a tail of the empirical distribution of the 1000 differences from the resamplings, then we have a choice to make:
(1) The two teachers did equally well but just by chance in the luck of the draw of the students one of the teachers seemed to do much better than the other one.
(2) The actual difference is so far out in the tail that we don't believe that the two teachers were equally good, reject the hypothesis that there was no difference, called the null hypothesis, and conclude that the teacher with the higher actual average was actually a better teacher.
Sure, it happened that the real reason was that one section of the course started at 7 AM and was over before the sun came up and the other section was at 11 AM when nearly everyone was awake. We like to f'get about such details! Or, sure, we might get criticized for a poorly controlled experiment.
This is also called a statistical hypothesis test or a two sample test. It is a distribution free test because we are making no assumptions about probability distributions of the student scores, etc. Since we are not assuming a probability distribution, we are not assuming a probability distribution with parameters and, thus, have a non-parametric test. Uh, an example of a probability distribution with parameters is the Gaussian where the parameters are mean and standard deviation.
Such tests go way back in statistics for the social sciences, e.g., educational statistics.
In more recent years, leaders in resampling include B. Efron and P. Diaconis, recently both at Stanford.
Why teach such stuff? Well, some parts of computer science are tweaking old multivariate statistics, especially regression analysis, and calling the results machine learning and/or artificial intelligence, putting out a lot of hype and getting a lot of attention, publicity, students, and maybe consulting gigs. Also the newsies get another source of shocking headlines to get eyeballs for the ad revenue -- write about AI and the old "take over the world ploy"!
So, maybe now some profs of applied statistics, what for a while was called mathematical sciences, etc., or other profs of applied math want to get in on the party. Maybe.
What can be done with resampling tests? I don't know that there is any significant market for such: Long ago I generalized such things to a curious multidimensional case and published the results in Information Sciences. The work was a big improvement on what we were doing in AI at IBM's Watson lab for zero day monitoring of high end server farms and networks. Still, I doubt that my paper has ever been applied.
One of the best areas for applied statistics is the testing of medical drugs. Maybe at times resampling plans have been useful there.
I have a conjecture that resampling plans are closely tied to the now classic result in mathematical statistics that order statistics are always sufficient statistics. Sufficient statistics is cute stuff, from the Radon-Nikodym theorem in measure theory and, in particular, from a 1940s paper of Halmos and Savage, then both at the University of Chicago. Some of the interest is that sample mean and sample variance are sufficient for Gaussian distributed data, and that means that, given such data, you can always do just as well in statistics with only the sample mean and sample variance and otherwise just throw away the data. IIRC E. Dynkin, student of Kolmogorov and Gel'fand, long at Cornell, has a paper that this result for the Gaussian is in a sense unstable: If the distribution is only approximately Gaussian, then the sufficiency claim does not hold.
Other applications of resampling, such applied math, etc. might be in US national security. E.g., maybe monitoring activities in North Korea and looking for significant changes ....
Maybe there would be applications in A/B testing in ad targeting, but I wouldn't hold my breath looking for a job offer to do such from a big ad firm.
For all I know, some Wall Street hedge fund or some Chicago commodities fund uses such statistics to look for significant changes in the markets or anomalies that might be exploited. I doubt it, but maybe! Once I showed my work in anomaly detection to some people at Morgan Stanley, back before the 2008 crash of The Big Short, and there was some interest for monitoring their many Sun workstations but no interest for trading!
Net, IMHO for such applied math: If can find a serious application, that is, a serious problem where such applied math gives a powerful, valuable solution, the first good or much better solution, with a good barrier to entry, and cheap, fast, and easy to bring on-line and monetize, then be a company founder and go for it. But I wouldn't look for venture funding for such a project before had revenue significant and growing rapidly and no longer needed equity funding!
Otherwise look for job offers (1) in US national security, (2) medical research, (3) wherever else. But don't hold breath while waiting.
Now you may just have gotten enough from about 1/3rd of the Berkeley course!
We teach these methods to our students in intro stats at UC San Diego. Have been for as long as I've been here (5 years). Last year a data science program was also created here at UCSD. I've TA'd a flagship course in that program too. It's almost exactly the same content; the major difference is imo are the faculty personalities. The stats profs are smug, while the data science profs are energetically self-important. They teach the same shit. Self motivated students with a STEMy personality tend to learn more in the stats courses because the profs drive on hard core theory; on average though, students do better in the data science course because the profs are so bombastic the kids walk out of each class thinking they are basically ready to join the fellas over at Waymo on some machine learning projects - maybe even show 'em a thing or two, cutting edge tricks learned back at the ol' uni.
Yup. Thanks.
> known as the empirical distribution
Yup, and I wrote:
"out in a tail of the empirical distribution"
Yup, "rank" tests, "permutation" tests: With my TeX markup:
E.\ L.\ Lehmann, {\it Nonparametrics: Statistical Methods Based on Ranks,\/}
And, yup, again with my TeX markup,
Bradley Efron, {\it The Jackknife, the Bootstrap, and Other Resampling Plans,\/}
Last time I knew, Roger Wets was at UCSD. He read one of my papers and suggested JOTA where I did publish it!
I'm a pure CS / logician by training, but I've spent a few years trying to expand my expertise into probability theory and stochastic processes. Lots of your advice resonates with me. My MSc advisor recommended I should go through Neveu. He was pretty good, had been a student of Pontryagin.
Neveu is elegant beyond belief. I keep my copy close. I was aimed at Neveu by a star student of E. Cinlar, long at Princeton and before that at Northwestern -- long editor in chief of Mathematics of Operations Research. Neveu was a student of M. Loeve at Berkeley. So was the current darling of machine learning, L. Breiman, because of his Classification and Regression Trees (CART). Breiman's Probability as published by SIAM is generally easier reading than Neveu.
For stochastic processes, there are several relatively different directions to go.
Martingale theory is gorgeous, astounding, amazing, with one of the most powerful inequalities in math, the astounding, tough to believe, martingale convergence theorem, and likely the shortest proof of the strong law of large numbers.
Then can do Markov processes more generally. The discrete state space version is important and not too difficult -- Cinlar has a nice introductory text.
A high end direction for Markov processes is potential theory. There are claims that that is the math for exotic options on Wall Street, but I doubt that there have ever been any applications.
There is a big role for second order stationary stochastic processes in electronic engineering. I ran into that for processing ocean wave data for the US Navy. Here the fast Fourier transform added a lot of interest.
And there's more.
Generally long Russia, France, and Japan seemed to have emphasized stochastic processes more than the US. But by now I suspect that the US is well caught up.
I'd have a tough time believing that very many people with money to hire know enough about high end stochastic processes, or even just Neveu, to hire in those fields. US national security may be about the only hope, that is, outside of academics.
Yes it appears that some of the quantum field theorists in physics are interested in path integrals.
Uh, I'm disorganized here: There is the field of stochastic optimal control!
As usual for advanced applied math, my suggestion is, outside of academics or US national security, find a valuable application and start a business to make money. That is, don't expect to be hired.
Incidentally, I have a copy of Loeve's old probability textbook - I wonder if it is too out of date to be any use.
Basically we're talking graduate probability based heavily on measure theory and some on functional analysis.
Last I heard of Neveu he was back in France and at least in part working on probability for stock markets.
There are other more recent authors. It seems that a big fraction of the best researchers sometimes take out a year and write yet another book comparable with Neveu. Personally, I did enough with Loeve, Neveu, Breiman, and Chung and for more attention to such material want to use that background to move on. For now, I'm concentrating on my startup; it's based one some applied math I derived, but I've done that and programmed it. Tonight I got four hard disks installed in my first server. Three of them are partitioned and formatted, and as I type this the fourth is being partitioned and formatted.
Well, there is more code to write, but IMHO that would be for relatively advanced techniques or, say, working with terabytes of data instead of megabytes.
If you want to write code for applied statistics, then maybe so indicate, have a portfolio of code, and contact the usual suspects -- US national security and medical research. I'm not optimistic. I've given my opinion -- find a good application and found a startup to monetize it.
It is true that today there is a WSJ article on how technical, with algorithms for trading, Wall Street has become. The article has next to nothing on what applied math is being used but does have lots of names, maybe some you could contact. Actually, the article mentions that Goldman Sachs (GS) got hot on such applied math. Well, that was about when I wrote Fisher Black, of Black-Scholes, there at GS asking about applied math at GS, and I got back a nice letter from Black saying that he saw no such opportunities. Well, the WSJ article today claims that that time was when GS was getting hot on applied math.
If you want to know about applied math on Wall Street, then try to get an opinion or overview from, say, James Simons.
Again, IMHO, it's academics, US national security, medical research, maybe a few other situations, but best of all, start a business, the money making kind.
To be fair, resampling wasn't the key to our projects, but we were doing a lot of work understanding probability distributions which is not entirely unrelated.
The idea of resampling is just dirt simple; read it again, just the first paragraphs, and you'll get it.
But, this bring us back to a much more central topic in data science: the tools and environment DO matter. Hugely.
Reproducibility is central to not just data science but all science. This is facilitated by the use of Free, Open platforms which adhere to common standards.
Imagine trying to debug why someone has a different answer than you when there are x*variant-of-program environments in which they have obtained their answer?
At the most basic level this course should be distributing a Docker image or a VM image of some sort in order to ensure that everyone has the same version of the software.
Even if you do not care about any of the above, please, shed a tear for the student who would like a simple setup.
Thank you.
What's the difference between, say, a Master's program in Computer Science where one studies machine learning and a Master's program in Data Science? Am I wrong for thinking the Data Science program weaker?
Data Science and DevOps are both just labels for things people have been doing under more mundane terms for 40-odd years.
Even Machine Learning is just a trendy buzzword for what used to be called Predictive Statistics.
I've never seen any stats text book or course discuss techniques for dealing with large amounts of data to any significant level, but in data science that is a core part of what you do.
I ran production systems before DevOps and after. Again, it's very different - prior to devops, there was no emphasis at all about using software engineering techniques to manage and deploy software. The most you'd get was some scripts maybe kept in source control if you were lucky.
Now I run an AI company, and a key part of the ML we use involves generating structured text files from images. I guess predictive statistics is technically a correct label, but the tools and techniques are so dramatically different that that thinking of them as separate fields is more correct than incorrect.
https://www.edx.org/xseries/data-science-engineering-apacher...
Go to community college. It's ridiculously cheap, and the credits are worth something.
Secondly, a data science course may be offered, but only during fall semester and everyone else wants to sign up for it.
Also bootcamps can compress the curriculum from two years worth of junior college into 8-12 weeks. For someone who never enjoyed being in school, I'll take the bootcamp.
Right, and I got a doctorate by listening to an audio tape.
It seems very short sighted; community college only takes 2 years. Your career lifetime is what, 45 years? The ROI is insane, why would you shortchange yourself?
The only boot camp grad I’ve spoken to personally had a degree in Human Genetics, did App Academy after deciding they didn’t want to do it as a career and got a Django job out of it, despite having no knowledge of Python. They had learned enough Ruby on Rails in three months of more than full time work to impress in an interview.
Community college is only two years? That’s half the time needed for an actual Bachelor’s degree, but radically less valuable than one for getting a job. There are two sensible reasons to get an A.A. or A.S., to get a job afterwards or to get the Bachelor’s that comes after. You need to pay for it but more importantly you can’t get a real job during it and you need to eat and live during it.
Even if a good boot camp is strictly inferior to a median A.S. in Computer Science the first can still be a better choice purely because it takes less time. Having known people with Bachelor’s in CS who can’t code I doubt an Associate’s is better.
Who would you hire? The boot camper with two and a half years work experience or the A.S. graduate with one? What if neither of them has a B.A. to go with it? What if both do?
https://www.youtube.com/watch?v=xcgrnZay9Yc&list=PLFeJ2hV8Fy...