Slashdot Mirror
Ask Slashdot: Books for a Comp Sci Graduate Student?
peetm (781139) writes "Having visited with me and my wife recently, the girlfriend of an ex-student of mine (now taking an M.Sc. in pure CS) asked me to suggest useful books for her boyfriend: '... He recently mentioned that he would love to have a home library, like the one you have, with variety of good, useful and must-have books from different authors. ... Mostly, I was thinking your advice would be priceless when it comes to computer science related books, but .. I would appreciate any sort of advice on books from you. ...' Whilst I could scan my own library for ideas, I doubt that I'm really that 'current' with what's good, or whether my favorites would be appropriate: I've not taught on the M.Sc. course for a while, and in some cases, and just given their price, I shouldn't really recommend such books that are just pet loves of mine — especially to someone who doesn't know whether they'd even be useful.
And, before you ask: YES, we do have a reading list, but given that he'll receive this as part of this course requirement anyway, I'd like to tease readers to suggest good reads around the periphery of the subject." I'll throw out Pierce's Types and Programming Languages (and probably Advanced Topics in Types and Programming Languages ), and Okasaki's Purely Functional Data Structures .
And, before you ask: YES, we do have a reading list, but given that he'll receive this as part of this course requirement anyway, I'd like to tease readers to suggest good reads around the periphery of the subject." I'll throw out Pierce's Types and Programming Languages (and probably Advanced Topics in Types and Programming Languages ), and Okasaki's Purely Functional Data Structures .
Code Complete has been one of the most useful texts that I have found.
http://www.edwardtufte.com/tuf...
They're kind of dated, because few people do sorts and list manipulation at that level any more. I have both an original edition and a current edition of vols. 1-3, but haven't looked at them in years.
Sure, for the average programmer these days who relies on existing libraries, these probably aren't all that useful.
As a grad student working on a thesis and other papers however, Knuth's books are invaluable for citations. Need to defend the use of a specific algorithm? Cite Knuth. His books were invaluable citation material for when I wrote and defended my thesis a few years back.
This is, of course, good science. You may not need to use Knuth to program your own B* tree, but you have a pretty much universally accepted reference for citation if you use one in your research.
Yaz
You seem to misunderstand what a citation is. A citation is not saying, "Here, this is true because Mr. XYZ said so," but, "Here, this is being accepted as true (or just considered even) because Mr. XYZ gave good argument/data in source ABC." It is not an appeal to authority, it is reducing redundancy by not re-deriving something already written about elsewhere. And this is a good science practice, as it is rare to be working on a topic where no one has already done some theory work or found some relevant experimental data.
You should have spent a lot of time deriving things in your undergrad and early grad courses, and you should double check the calculations of any citation central to your work to catch typos or actual mistakes. But by the time of writing actual papers or your doctoral thesis, you shouldn't be wasting reader's times by re-deriving things in detail beyond what is needed for outlining background and covering your premises. Unless your new contributions are a better way to present or derive the same solution, or showing a mistake in the derivation, you move on to deriving new things while cite others for stuff already done.
Cite Knuth... This is, of course, good science.
Well at least Professor Knuth is still alive, and I don't [YET!] need to refer to the poor man as spinning in his grave.
AC posted an excellent response here.. In the event you're filtering AC's, take the time to read it, as it's completely on point.
I would add is this: if you've never completed a Masters thesis or Doctoral dissertation, just try submitting one to your committee without adequate citations. If you write somewhere "I used well-known algorithm ABC because of XYZ" and you don't have a citation for that algorithm, you'll be sent back for rewrites pretty quickly to add appropriate citations.
By way of example, in my Masters thesis several years ago, I mentioned Unix diff , without a citation. Why would this need a citation? It was mostly mentioned in passing, and every computer scientist under the sun knows what diff is, right?
Committee came back asking for further citations on a few things, including diff (which, for the record, is "Hunt, J. W., and McIlroy, M. D. An algorithm for differential file comparison. CSTR, 41 (1976).")
Using citations isn't an appeal to authority. It's akin to using an existing library call in programming. Just as you wouldn't roll-your-own quick sort algorithm when coding, someone writing a scientific paper doesn't re-invent every algorithm ever derived. You find someone who has already done that, and you cite them. The AOCP is useful in this regard due to the sheerly massive number of algorithms Knuth describes. It's hard to go through a Computer Science program and not use one of these algorithms. Knuth himself likewise cites all of the algorithms in the AOCP, so it's not an appeal to his authority, as he delegates that out to others appropriately. It's simply useful because instead of having to track down papers written in the 1960's on your own, you can cite Knuth who cites those papers for you. This is why the AOCP is useful for a graduate student.
FWIW, I cited Knuth. I needs an algorithm to calculate variance, and another on the Box-Meuller transformation. Art of Computer Programming had one for each, which I adapted for my needs, and cited appropriately.
Yaz