For me at least, sparse vector support means you can do elementwise operations (on the non-sparse elements) and in particular linear algebra like vector dot-products and matrix-vector multiply.
I have been thinking for a while now about the applications of bandits to financial markets (not so much the Q- and TD- learning approaches as I am a bit less familiar with them).
I will definitely have a detailed look at the thesis, sounds very interesting!
+1 for Langford. He and many others (e.g. Deepak Agarwal) at Y! are among the most prolific publishers on this topic. Check out: http://hunch.net/~exploration_learning/ for a good, but pretty technical overview.
Sofia-ml which is a very fast linear svm and classification C++ package. Supports PEGASOS as well as logistic regression and also learning rankings. Has no bindings for other languages which is a bit of a downside. Still, a useful command-line tool.
It also includes a package for very fast mini-batch K-Means (http://code.google.com/p/sofia-ml/wiki/SofiaKMeans). Combining these two approaches one can effectively learn a "kernelized" model while still being linear and therefore very fast (at least this is the claim, I haven't tried this).
I've used both the SVM and k-means package and they work very well. For sparse datasets with >500 dimensions and > 10 million rows, file IO time was <15 sec, training time <3 sec. K-means is slower but still orders of magnitude faster than standard batch k-means.
Finally, Vowpal Wabbit is a very fast package that also uses stochastic gradient descent as the workhorse. Also has a nice feature-hashing compression scheme which is being widely adopted (e.g. in Mahout, and also in sofia-ml above).
I'm curious as to what libraries are available for linear algebra and numerical computation (free / open source ones) and how they compare to e.g. Numpy, mpj or colt on java, etc?
I wonder what results a study into the differences in brain function (eg fMRI scans etc) between recalling normal memories and "fake" memories, perhaps using machine learning, might turn up? I wonder if such techniques could distinguish between real memories and fake repressed memories...
I think Mark Shuttleworth's Thawte (bought by Verisign) is the only real example of a startup in the sense most on HN would think about it, at least at scale.
It's also worth noting he started Canonical which gave us Ubuntu.
But in South Africa at least, there are plenty of startups - again a much smaller scale than US or Europe - in software and Internet. Few people even in SA know about them, so I wouldn't expect pg to leap in to fund a bunch. Africa needs to develop it's own pg's. I actually think the approach of "micro-angel-seed-VC" is a good fit for the funding problem.
The fact that Dubai salaries are tax-free has a big impact too (although cost of living is quite high too (not relative to big US or UK cities but pretty high)).
Also depending on the gig often relocation expenses and other expat benefits are a factor (probably more so in the financial firms and big cos though).
Hopefully! It is great to work with some real world customer click data...
If anyone is interested in a technical intro to the setting there is a set of slides from John Langford at Yahoo Research (many good and standard reference papers cited in it): http://hunch.net/~exploration_learning/
A/B testing could be thought of as a sort of epsilon-Greedy strategy (particularly if such testing is carried out at regular intervals initially). While not enjoying the optimality characteristics of other algorithms, such an approach can in fact outperform in many practical cases :)
Masters project on using bandit algorithms for optimising CTRs on website content. Also involves some search engine / text mining / dimensionality reduction stuff.
A little bit of messing around with Android SDK too...
Only very few people have commented that, you are only 25! (I guess given the nature of the HN community :)
I don't think anyone can consider themselves a failure at 25, having spent 8 years learning a lot of programming, algorithms, data mining / machine learning and complex systems modelling.
It would be easier to think that if you hated what you were doing. But it sure doesn't sound like you do. So you just need to decide what you really want to be doing.
If it's algo trading, I concur with another poster that says go work for a hedge fund (or bank, or prop trading shop). It is super competitive, but they have the technology infrastructure and most importantly the capital. Getting in is not easy, but simply show them all your work (it doesn't work anyway, but is indicative of your skills and way of thinking). You will learn a lot, you may hate the people and environment, or you may love it even more. And then yes after a few years of experience you will most certainly be in a better position to go off on your own again (or do something totally different, by then you will really know if you like it or not!). Many top hedge fund managers / traders only started their own thing at 30, 35, 40, even 50... I know a dentist who became a prop trader. Anything is possible.
A PhD would be a great option IF it's for the right reasons. But if you want to do a startup (sounds like you might quite like the idea and you posted here on HN, so...):
- you already live on ramen, so no lifestyle change there;
- bootstrapping something can't cost more than losing money with trading algorithms;
- you already have many of the requirements: coding / technical skills, low-cost living circumstance, love to solve tough problems and a huge amount of tenacity in the face of failure and overwhelming odds;
- bonus: your interest in social systems modelling etc ties in pretty nicely with what's big right now and in the near future.
So if that is what you really want to do, go for it either alone (or find a co-founder), or find a small/medium startup to work for. To make the transition a bit more natural perhaps focus on ones that are data-driven and have machine learning / modelling at the core of their business. Think recommendations, systems modelling (www.flightcaster.com) and weather (www.weatherbill.com). There are many many examples of YC and other startup companies of this nature (many focused on the social network space).
Good luck in whatever you do decide to do with the next 60+ years of your life. On your deathbed you can post about whether you think you are a complete failure or not.
What's interesting about this is the mention of how they built a new and improved software system for the cash management. Wouldnt be surprising to see this pop up as a future cloud-based product, as part of their google docs offering.
I am interested in this, but is it worth going if not a developer/designer but a (soon to be) machine learning masters graduate? ie does one get anything really out of it if you can't contribute much in the way of actual code,design etc? Anyone with similar background attended and found it good?
This article is at best inconsistent, at worst self-serving.
Funny that Cuban wasn't complaining about traders when they were driving up tech stocks in the dot-com bubble that eventually helped him land his giant payout in Yahoo stock. Or the traders that took said Yahoo stock off of him when he wanted to divest himself (aka liquidity).
Tax breaks for long-term investors? Well VC and PE firms are "long-term investors" and they already have a pretty sweet deal on carried interest. That didn't stop them from rampant speculation in the tech bubble, or buying companies at outrageous debt multiples in the credit bubble. Imagine how much more it would have gone on if they didn't have to factor tax into their return calculations? I bet Mark would love to get all his future speculative VC investment payouts tax free. I don't know of a major tax regime that doesn't treat short-term trading revenue as income as opposed to capital gains. Tax breaks for "real long-term investment"? Yes, it's called a 401k or pension plan. If you're really a long-term investor a 10% algo glitch (4% actual down day) shouldn't even register for you. You shouldn't even be looking at the market on more than an annual basis.
"The market has changed over the last 3 years" and is driven by macro issues? Yes we've been through a severe global recession driven by the bursting of a global credit bubble. Of course macro issues have dominated and volatility has been extremely high.
(The wider) Wall Street's business is not and never was purely raising capital for companies. In the 20s were market volumes way less than capital raised as a proportion? I doubt it, but even if so it sure didn't prevent the Great Crash. Why does government need to further incentivise capital raising, that already brings in some of the highest possible fees (on individual transactions) to investment banks of any "traditional" activity (equity and debt underwriting, M&A, market making), i.e. excluding principal business and super exotic trades. He also says that companies never go public anymore, yet global IPO volumes are the highest in a decade (http://www.efinancialnews.com/story/2010-03-26/global-ipo-q1).
The only point I really agree with is leverage - it fuels the growth of bubbles, and exacerbates their bursting (or causes it when taken away in some cases). Most of the major bubbles/crashes in the past 100 years were either caused by excessive leverage (credit bubble, LTCM, various debt and currency crises) or had leverage as a major feature in their bursting and the speed and volatility of the movements (dot-com, Great Crash, credit bubble). The possible exception to this is '87 (to a large extent computer-driven, arguably) although again leverage played a big role.
Finally, the major, unforeseen crashes have typically come about due to opaque, illiquid and/or highly leveraged situations (AIG, LTCM, CDS on CDOs, day traders buying tech stocks on margin etc). High-frequency traders only trade in the most liquid instruments, so ironically they are the ones trading on exchange, transparently and with exchange-set collateral and trading rules. Certainly illegal activities such as real front-running should be stamped out. Possibly things like flash orders too. A level playing field should be ensured. But he should worry more about the stuff going on off exchange than in the public markets.