east2west

I suspect money is a big motivation behind all the hand-wringing by statisticians. It is about jobs, grants, prestige. I don't blame them; people want these things. I defintively see a competition between computer scientists and statisticians for the mantle of date scientists. When the current bubble bursts and the hype dies down, one side could have won and the world could be very different, or not. What do I know?

I am a little torn between realistic business demand of paying least money to get jobs done and thorough mastery and understanding of theories. Look at biostatistics; there master graduates become data analysts and process data. They know the nomenclature and can understand instructions from more advanced statisticians. The majority of practitioners don't know proofs and they carry off their tasks reasonably well. I would call all but very select few statisticians applied statitisticans and the world is fine with that. I am not sure machine learning industry will not turn out like this, although I am not looking forward to that day.

I would just note that numerical optimization is a big deal in machine learning while I know of no statistical department strong in it. The popularity of SVM owes a load of gratitude to pracitical breakthrough in dynamic optimization. But if one elevates to the level of functional approximation then statistics is similar to almost all engineering fields. In fact, I suspect statisticians have more common with engineers than scientists. I bet Eric Cantor didn't care that polling results had margins of errors but that he lost his House seat due in part to overconfidence coming from unreliable statistics. Real world applicability, not theoretical nicety.

OK, to be fair, R is a reasonable language for statisticians and others. I guess R and S are different enough from usual programming languages that I may be biased against them. But I did spent five years working on bioinformatics and statistical genetics using it, and I couldn't wait to get away from it when I did. I will list my personal peeves.

1. R's syntax has problems. This is well documented on the web. I recall : has wrong precedence.

2. My biggest problem is R's lack of basic abstraction facility. Beyond functions there is nothing good. There are three object-oriented programming systems, none of which are simple or straightforward. It has first-class functions and lexical closure so at least basic functional programming should have been possible and fun. But R functions' have a weird parameter passing mechanism which scared me away from investigating deep. In the end I used RCPP to write large codebase and expose an R interface. And I like to delve deep into things. I was learning template metaprogramming when I was working on bioinformatics.

3. CRAN has way too many low quality packages, but far more importantly almost all packages' documentation consist of use-cases plus reference. To this day I don't have a good mental picture of genomicRange (a foundational bioinformatics package in Bioconductor for NGS data analysis) data structure and how to manipulate it (one main data structure, I believe). This is annoying at best and infuriating at best.

4. R, more than any other language I know, is an implementation-defined language. The language reference manual frankly admits fuzzy rules for some of the more advanced features. Python, which has been recently criticised for similar problems, is clear at the basic language features level.

I don't want to this discussion to degenerate into a flamewar. Programming language preference is a highly personal thing and I respect people who use R, even admire programmers who create awsome packages. ggplot2 and Plyr are some of the best packages I have ever used, and their documentation is superb. I just don't believe R is a good language for statistical students if they hope to move on to professional programming in other languages.

The following quote is from "Matrix analysis and applied linear algebra" by Carl D Meyer, p 423-424.

"Generalized inverses are useful in formulating theoretical statements as those above, ut, just as in the case of the ordinary inverse, generalized inverses are not practical computational tools. In addition to being computationally inefficient, serious numerical problems result from the fact that A+ needs not be a continuous function of the entries of A. ... Not only is A+(x) discontinuous in the sense that lim_{x->0}A+(x)!=A+(0), but it is discontinuous in the worst way because as A(x) comes closer to A(0) the matrix A+(x) moves farther away from A+(0)."

Here A+ denotes Moore-Penrose pseudo-inverse and A(x) = [1, 0; 0, x] in Matlab notation (or Numpy? I am very rusty on Matlab). And in case you are wondering why the generalized inverse is plural, there are different kinds of generalized inverses, Drazin inverse being a well defined alternative.

Talk about head in the sand. CS is subsuming stats (not really) because computers are a, if not the, major driving force begind statistical advances. L1 regulation became a fad in stats because computational optimization could efficiently solve this non-smooth optimization problem. Engineers are now advancing algorithms and moving it to real world use, while statisticians are still re-searching their field and finding applications using existing algorithms. There is nothing wrong with application, but this means that they are not at the forefront of research and more importantly they have to wait for other people to solve their problems because they are no capable of inventing novel algorithms. When the world at the large relies on computers for every little thing and on computer scientists for a lot of scientific advances, is it really surprising that computer scientists are more prominant?

And computer scientists have plenty of rigor. NIPs conferrence is an AI/neuroscience conference that is now prestigious in stats as well. Some machine learning experts lack mathematical foundations but statisticians are not really good mathematicians either. Being able to regurgitate stat theorems and prove convergence in expection under two dozen conditions (no kidding; read average JASA papers) are not the hallmark of analytic rigor. Average stats Ph.D., even from top programs, lack thorough understanding of measure theory. Just look at standard stat textbooks; examples are aplenty (using statistical software), but proofs are nowher to be found. It is amazing how many statisticians are still using pseudo-inverses in numeric computation.

The statistical education is a problem. The biggest factor being the very software this article is promoting: R. It is an atrocious monstrocity. Perhaps statisticians can live with it. Fine, but don't expect good programmers coming out of using R for five years. Python would be far better. In the end, statisticians are not professional programmers, the same way accountants are not. I don't see too many accountants fretting CS is subsuming their field. Let us not forget, most statistical methods have their roots in other science and engineering fields, and they are not particularly bothered.

Yes, but Jefferson's nephew in South America, let's say, post-Civil war hardly can be called well known or powerful. It is no given that every Roman noble had nothing to fear from pirates. Caesar's contemporary Claudius fared much worse when he fell into pirates' hand, even though he actually came from a great patrician clan and was related to a man no less than the prominent Sullan general Lucullus. This episode, if true, delineated a subtle, bold, and decisive nobleman who did not shirk from the responsibility of his rank. It is no accident he was called the last patrician.

As much as I hate that sceintific papers don't give good introductions to the public, specialists do have their values. An educated, intelligent reader has limits when it comes to frontline research. It is like asking someone flying a 787 jetliner without touching an airplane before. Prior knowledge and experience are important.

What I absolutely reject is false exclusivity that some scientists hide behind. A professional statistician can spot statistical errors even if he or she never did any cancer study before. A rational person can discern the trustworth of a scientific report if enough details and background are well explained. I always find that popular press lacks a degree of healthy comptempt for specialists. It actually has improved somewhat recently. Still sometimes newspaper reports read like press release.

This is a rather short piece covering big areas. I half expected her to explain her precise role and her projects at Intel after reading the first paragraph. She is a good publicist, the way she pivots questions.

This covers all issues surrounding reproducability pretty well. It is astounding that scientists want to hide behind their expertises when their results and conclusions are being tested. Biologists who are under a similar cloud tried the same tactics too. They have some secret sauce that make their lab and experiment special and unique, beyond the comprehension of mere mortals. Of course, as publically funded scientists they have a moral if not legal obligation to disclose their "secret sauce" if that is very important. Why publish a paper but leave the public in dark about the most important factor? At best it is an oversight, at worst it is dishonst.

Experiment and reason underpin the modern science. Common sense says inreproducible experiments call their colusions into question. Scientific experise is no escape from reason. A fancy degree from a "top" university does not make infallible anyone. Well controlled experiments are the final arbitrator.

If multiply operator weren't such a big issue, the creators of Numpy wouldn't have attempted to insert a new operator for matrix multiplication into Python. For a dynamic language such as Python operator overloading for such closely related types is a big trouble. If I write a function that uses multiplication, either I use member methods such as "dot" or check for type explicitly, otherwise there is no guarantee what would happen. The worst part is that errors are strictly logic and only way to debug is to trace from end result all the way up to the point of object creation; it isn't pretty.

You know academy is not infallible and scientists are no saints. And boy, are scientists no saints. Passionate about science, please.

Go to Science Career forum (hosted by Science magazine) and see how urgent the need is to cull Ph.D. crowd.

Modern war is more destructive than previous ones. Even fifty years ago genocide was a laborious task, now it is done with a press of button. There is nothing more deadly than total war.

I understand where you are coming from, but your facts are not not all correct.

> I no longer die from the flu and most of the common diseases have been all but wiped out. Hundreds of thousands of people die from seasonal flu every year, pandemic flu can kill 5% of population if let loose. SARS and the recent H7N1 flu all have pandemic potential, not realized only because of massive social response at patient isolation. Not only are we far from wiping out common diseases, a lot of them is making a comeback. Some scientists believe we are on the brink of post-antibiotics era. > People are using binary logic to program bacteria to kill disease. That is just not true. What synthetic biologists claim to be binary logic is basic common sense redefined. And the only reason they use bacteria is because they would be completely lost with an Eukaryotic cell. One should not mistake grant-hunting PR for reality. > If any organ in my body has a problem it can be replaced with a donor version. Many can be replaced with artificial versions, sometimes printed to my exact specifications. You are probably wording this wrong. No one is replacing brain for instance. And from my corner of world, considering how little we know about the inner working of the immune system, no way all organs can be replaced to individual specification. In fact I don't know any ORGAN can be fabricated to individual specs. There might be, but I would be staggered. > When was the last time in living memory that a person could read most works of science fiction and say "Yeah, with the right kind of funding, we can have that in 10 years"

> Stop lamenting the lack of progress and look around. The world is completely different than it was even five years ago. I don't think most science fictions are picturing a technical utopia nowdays. If anything, there is more than a little hand-wringing. Progress can look different to different people, and we are no longer so naive as to believe the unalloyed good of technology.

I share your belief progress is being made, but as always, haltingly, especially for their social impact. People here are generally technologists and want to believe in the intrinsic value of technology. But I bethink myself and remember Goethe's "Faust" and its caution against head-strong rush toward power and "progress."

I wonder if anyone knows why subtlety exists in Python. Is it because reference semantics and HOF are fundamentally incompatible, or could it have been avoided somehow but still retaining reference semantics. The most common complain about lambda in Python is that it is restricted to expressions, which is a conscious design decision that could have gone the opposite direction.

I have been eying F# for bioinformatics work for a year, but a dearth of bioinformatics library, difficulty linking existing libraries, and flaky tools on Linus have held me back. Right now I am stuck with Python and R, which are good languages with plenty of good libraries, but I want to branch out into functional languages. It also does not help that I don't know C# and am disinclined to learn it.

HN is understandably filled with programmers who value programming above all else and believe domain knowledge can be picked up quickly. It would be true for someone hired as a programmer who implement a published algorithm or port existing software to newer systems or different platform; in other words, improving existing systems. I believe it requires equal depth if not not more for domain specific knowledge to innovate in non-computational fields. For my field of bioinformatics/computational biology, many open questions cannot be answered because of computational hurdles. Reading textbooks won't lead anyone to them, nor would reading survey papers. It requires expert knowledge in both biology and computer science to even know what is feasible and interesting. We want to find the driver mutations for cancer, discover mechanism of gene regulation, uncover subtle interaction of heredity and environment. Computation is intimately involved because observation is indirect, noisy, and massive. Only with knowledge in both fields can answer questions that domain experts care about but cannot do because of computational/quantitative obstacles.

I find researchers (I am in academia) in both fields cross-talk too much. Computational/quantitative experts try to shoe-horn biological questions into their small corner of research area, while biologists don't care about computational efficiency, statistical robustness/efficiency, and the mathematical implication of their simple methods; no one cares about software quality. This is unfortunate. It is also unfortunate hiring committees do not look favorably towards multi-disciplinary researchers because one is not a trained geneticist/cancer researcher. I definitively do not encourage computer scientists go into support roles in academia; they are poorly paid and get little respect.

A few things in this pierce struck a cord with me. I am a computer science Ph.D. doing a postdoc in statistics department, working on biostatistics, and as I am recently looking for a job and did a few interviews, some observations may be helpful to job candidates.

First, people are pigeonholed into certain roles. Statisticians analyze data and design experiments. They are expected to know classical statistical fields, theorems, and their standard usage and caveats. Programmers should be able to answer standard programming questions, such as dynamic programming, balanced tree, etc., and of course write programs. In large company where specialization is the mantra, expertise in both fields is not an assert in interviews.

Second, multidisciplinary experience can be a liability. Since my experience in statistics is nonstandard, more of signal processing than classical statistical inference, I am not as conversant in classical statistical theory as a statistical Ph.D. does, especially because genomic research often prefers most basic methods, as in industry. Interviewers rightly ask about things they know, and are not impressed if one cannot answer questions they learned in graduate class. It is similar with computer engineering interviewers. They will ask me to implement an interval tree or dynamic programming algorithm in 30 minutes. Most of my heavy programming is in numerical analysis and optimization, where dynamic programming is very different from what a programmer thinks it is.

Third, depth is not required in industry, and certainly not in interview. Interviews now-days feel very much like college entrance exam in China or one of those East Asian countries, where people are expected to regurgitate set answers and the most important trick is to meet expectation. It is not important to master materials but to have right answers. And the right answer depends on the person asking question. One may considers SVM to be mainly a kernel trick that molds nonlinear relation into linear function, while another considers SVM as finite approximation of dynamic optimization with a breakthrough in quadratic programming that efficiently solves the two-points boundary problem coming out of dynamic optimization. This gets back to pigeonholing roles. A professional statistician will prefer one while a control/dynamic system expert will like the other. The killer is that some interviewers ask questions with their preferred answer in mind, and the questions can baffling to people with different background.

Forth, different companies demand different capabilities for supposedly same roles. Data scientists can be as mundane as denormalization or as sophisticated as inventing a way of causal inference. It is not always easy to tell from ads. It is even less so when the company wants jack of all trades and experts in all possible tasks. Ask very pointed questions.

Fifth, there is no advantage in being both good programmers and good statisticians, at least in interviews. I have already noted several disadvantages. People much prefer build inter-disciplinary teams each member of which is tasked with one special area and let them talk. It works well. I cannot think of anything that requires expertise in both programming and statistics in one head. It may be a little slower, but not noticeable.

I am having doubts about a position in a large company because of too much specialization. I like to derive an algorithm and implement it efficiently. Even if I could get a position, and I could if I cram for interviews, I would be at a disadvantage when others can concentrate on one area. Sadly, academia is becoming very much like industry, except they count papers or grants instead of make money. I am still looking for my niche.

I guess the gist of my rant is that today's job market demands specialization and people better conform.

Did you mean Roman public education only taught basic literacy? For aristocrats, pupils were taught Greek and Greek literature early on since later republic era. Julia Caesar's mother went so far as to have a Greek slave accompanying Caesar from early childhood so he grew up bilingual. Marcus Aurelius was as comfortable in Greek as he was in Latin. His extensive training rhetorics is obvious from his writing. He wrote "Meditations" in Greek, because that was the language of philosophy.

samtools is among the better software in sequencing-data analysis. It is also great in defining formally the file format it uses. The same cannot be said of many other tools. For my recent work in RNA-Seq, samtools is the most robust and trustworthy tool I looked at and used, and I looked at almost every popular tool, the major exception being RSEM. If only all bioinformatics tools are more like samtools.

Nothing. It is bad to not parse files but still changing content because you assume the file format mandates something when it doesn't.

Interestingly, the fraudsters were caught because of a false claim on a CV, and that finally destroyed their creditability.

It is intentional fraud, no doubt about it; they restarted halted clinical trials. I was just pointing out they did sloppy work too.

That was a tongue-in-cheek thingy. I write extensive tests for all my code. But when I look for a job, people counts papers not weight software quality. It is not easy.

I wasn't appalled by AWK language but the blind switching of two lines which were assumed to be paired reads when many reads are not paired. It is precisely lack of checking that is a problem. I have nothing against AWK, although personally I use Python for massaging data.

It was initially a programming error but the Duke researchers refused to acknowledge it and reanalyze their data, because that might mean retracting their prominent paper. From there it just snowballed. It was certainly fraud after the error was pointed out to them. There might also be other elements of fraud in their paper. I watched a presentation by MD. Anderson researchers who spotted the error and spent more than two years trying to call attention to it.

No kidding about reinventing wheels. I once saw a manuscript based entirely on dot-product as 1-D least-square. I don't know what happened to it, but one reviewer called it a seminal event in GWAS.

Bioinformatics is hard, but too many careerists take advantage of difficulties and uncertainty to publish as many papers as they can get away with.

Well, I can call myself a bioinformatics researcher, I guess, as I have CS Ph.D working in genetics/genomics. I see your point of throwing computers at simple solutions as cheaper than throwing good programmers. I do that too. We are very fortunate in that we write run-once programs that only have to work in one environment using one inputs. However, bad programmers write incorrect programs, which give wrong conclusions that lead to faulty clinical trials (look up Duke University facing class-action law-suit). I have seen people parsing Gigabytes-files with one line of Awk. People seem to forget that good engineering practice is learned with blood. Is it any wonder academic research is looked with suspicion by the pharmaceutical companies?

Well, considering that Amgen could only reproduce 11% of "landmark" papers in cancer research, peer review is missing a lot. See "Drug development: Raise standards for preclinical cancer research" http://www.nature.com/nature/journal/v483/n7391/full/483531a...

And some of the scientists insisted on non-disclosure agreement before agreeing to help Amgen reproduce published results, which proves most papers don't publish nearly enough details to reproduce main results.

Peer review is a social organization that necessarily never lives up to its ideals. At top journals such as Science, Nature, Cell, editors also consider PR values, popularity of topics, exclusivity. It is standard practice to obtain prior favorable impression by contacting editors before formally submitting manuscripts at these places. There are different standards of peer review; consortium such as ENCODE and TCGA have a different track for their papers.

I don't want to get into a pissing contest but just to point out there are people equal of him in the versatility and greatness, such as his great rival Michelangelo, who is also a great sculptor and great architect. Da Vinci's work as a field engineer impugns his character as a shiftless ingrate who laid siege to his home city that did nothing but adoring and honoring him. Compared to Michelangelo, who grew up poor and labored under severe constraints, Da Vinci had a much easier life.

I don't see such a sharp divergence as a computer science major working in a top statistics department. Statistics does not magically work well in hard sciences and machine learning has worked very well in many fields. I don't know if you consider biology hard science but biostatistics has many pitfalls many sciences fall into, and highly regarded statisticians can come to opposite conclusions on same data.

Statistical theory is actually applied probability, often called mathematical statistics. Math majors would feel right at home. In fact math majors look down on statistics major for lack of rigor and they have a point. Two years of classes in a five years phd program isn't enough to perfect knowledge of measure theory. But the more important problem is that stats training leave out computing enough that it impacts works coming out of stats department.

This gets back to difference between machine learning and statistics. Machine learning research embraces all fields of engineering, approximation and estimation, numerical analysis, optimization, plus statistics. Since so much of scientific advances can be attributed to computational improvements, it is natural that the more computational oriented fields are ahead of less computational fields. LASSO has been all the rage in statistics recently when it largely relied on works in convex programming. And signaling processing community in EE and CS are leagues ahead of statisticians in the sophistication and scales of problems they can tackle. Computational statistics is an attempt to remedy computational shortcoming of traditional statistics, but we have yet to see visible impact in term of high-impact work from statistics departments.

Having said all that, computer science departments do have the problem of not fully understanding the statistical foundation of machine learning methods. But this is not the case for CS in prestigious schools such as Berkeley and Stanford and MIT. Work coming out of these places are theoretically sound yet application oriented. One needs to just look at NIPs papers to appreciate the breadth and depth of expertise available in machine community.

For a good reading on bridging statistics and machine learning, read paper by an inventor of random forest, Leo Breiman, "Statistical Modeling: The Two Cultures." It is a well regarded paper by a renown probabilist and statistician who cares about utility of statistics as used in the real world.

You make it sound like MMA is better than Matlab for even numerical work. Is this true? For example, would machine learning algorithms be better code in MMA than Matlab? Matlab right now dominates engineering and in particular machine learning.

I wonder how this will impact air travel. It is a real shame that after Concord retired there is no supersonic jetliner in service, which at least kept the dream alive even if Concord is expensive and short legged. I believe no viable program exists right now for supersonic jetliners. The problem, if I recall correctly from my aerospace professor, is not the jet engine's thrust but heat generated from friction with air. If this Sabre engine can get the airplane mostly out of atmosphere cheaply, heat problem can be alleviated. I don't know how practical my guess is, but if it works, the market potential is immense.