Three jolly sailors from Blaydon-on-Tyne They went to sea in a bottle by Klein. Since the sea was entirely inside the hull The scenery seen was exceedingly dull.
From the hard-to-find Space Child's Mother Goose by Frederick Winsor. This is a hilarious collection of nursery rhymes for scientists.
The classes *I* took used "NP-complete", "NP-hard" and "hard for NP" synonymously.
NP-complete is a subset of NP-hard.
NP-complete problems are all as hard as each other in the sense that either all of them or none of them can be solved by a polynomial-time algorithm.
NP-hard problems are at least as hard as NP-complete problems. So, for example, the halting problem is NP-hard, but it is not NP-complete.
Anyway, this is not the part of NP-completeness theory that I think programmers need to know. What they need to be able to do is recognize that, for example, the way they are attempting to optimally put outgoing mail messages into the proper queue is equivalent to bin-packing, and therefore takes exponential time in the worst case. They can then go find one of the excellent bin-packing heuristics that has been developed, implement it, and then move on.
That's not the problem. It's not academics, really. Rather, the fault lies with those zealots who claim that anything running a Linux kernel Linux, as though that were all that mattered.... Sometimes they're just repeating what they've heard, not understanding that "distro" is a cutsie dodge to avoid saying "operating system".
Yes - sometimes the problem is ignorance, and sometimes it's zealotry. You might say that a zealot is just the kind of person who takes any difference ("there's only one Linux kernel") and thinks that it's an argument in favor of his favorite system.
If those steps could happen, then perhaps progress can be made.
Absolutely. It is important to recognize the inherent tension between vendors trying to differentiate themselves and vendors avoiding scaring off application developers due to the difficulty of targeting (in effect) multiple platforms. This conflict creates a tough problem.
I don't think that they will be, however, because too many people have too much ego wrapped up in the myth.
I don't follow you here. Which academics "repeat their myth about linux=o/s=kernel" and why do they do it? I'm not disagreeing with you - I just have not heard many people making this claim.
The most recent issue of Jakob Nielsen's excellent Alertbox column about web usability is about reputation management. It even mentions slashdot!
Re:This is called collaborative filtering...
on
Moderation Ideas
·
· Score: 1
Not quite...
I was proposing that my value function for/. posts would be calculated only on the basis of my past agreement/disagreement with posters and moderators.
If the server compares this profile with other people's profiles, and uses these to augment my profile, then it's collaborative filtering.
Currently, everybody sees posts that are ranked on the same scale: what the moderators think. I don't agree that this is (always) the right thing to do.
Rather, each reader should have the ability to see a customized view of/. based on their preferences.
For example, if I see a post moderated up to "5" and I think's it's a bonehead post, I would like to de-value the opinions of the moderators who liked that post. This is different from meta-moderation: I'm not saying that these moderators are bad and should be prevented from moderating, I'm saying that I don't want to see the results of their moderations.
Similarly, when I read/. (and the linux kernel dev list, and comp.lang.perl.misc) there are some people whose posts I consistently value. I would like a way to always read their posts. Conversely, I would like a/. "killfile" so that I don't have to read posts by people who have irritated me in the past. Again, I'm not talking about reducing their karma so they can't post - I want a filter so I don't see their posts.
Eventually, each user will build up a profile of posters and moderators whose opions they trust and distrust. One can picture people trading these profiles, merging profiles with other people's, etc.
As the amount of information on the net grows, developing a systematic way to figure out which of it you trust and which of it you don't is going to become a very important problem.
Didn't Brin have some sort of reputation management system in Earth? Reading up on this might be a good place to start.
Anyway, the point I want to make is the people have different tastes and want to read different subsets of all posts. There's no reason to force the same view of moderation on everyone.
When you look at all these factors, a price increase of $10 for hardcover over paperback is fairly typical. I'd imagine that you could get by with a $5 spread, but you'd be taking extra risk for benefits that were passed along entirely to the consumer.
An extra $10 for hardcover certainly seems reasonable when we're talking about a $40 softcover, especially since the online bookstores give a slightly larger discount on the hardcover.
So how would O'Reilly go about deciding which books (if any) are worth releasing in hardcover? It's not entirely clear to me that you could make a reliable prediction about something like this through a survey. Maybe it would be obvious just by looking at the 5 or 10 best sellers.
Publishing A book which sells for $40 can be produced at a marginal cost of $2. This gap between price and marginal cost has led to a variety of forms of differential pricing. Book clubs, hardcover and paperback editions, and remaindered books are all examples of the ways that the product characteristics and adjusted to support differential pricing.
here. That doesn't make it true, but it sounds reasonable.
"Price discrimination" and "differential pricing" seem to be the keywords.
I was thinking that O'Reilly could attempt to build (even more) customer loyalty by not trying to use price discrimination.
... Despite polls that said people would pay more for a more durable hardcover, everyone bought the softcover to save the difference in price.) So, if there is a poll, how much would you pay for a more durable book?...
I've heard that the hardcover and softcover versions of a book cost about the same to produce, and that the purpose of a hardcover edition is to extract extra money from people willing to pay more to read a book earlier (since the hardcover comes out well in advance of the softcover).
If this is true, how about releasing hardcover and softcover editions of O'Reilly books at around the same time for around the same price? Seems like it should increase sales and customer satisfaction. I'm definitely in need of a hardcover "Programming Perl"...
As another poster said, you will want to go to a school that has a group that is doing interesting work in an area that you're excited about. This is true, but there are a lot of other factors.
It is important to not only be excited about the reserach, but also to get along with the professor(s) who run the group. Many professors are not good managers - it's important to realize this early on. A big department can be an advantage since there's a better chance that there are other profs doing research that you're interested in if you decide that you can't work with the advisor with whom you originally wanted to work.
If you have two years of undergrad study left, there is a lot you can do to help figure out what you want to do in grad school, and to increase your chances of getting in. First of all, try to take a couple of graduate level CS classes in areas that you're interested in. See if you like them, and if you're comfortable with the workload. Second, you should definitely try to work for a professor in your current department. There are two ways that undergrads can be compensated for their work: money and credit hours. You will end up doing grunt work for a research group, but it's a great way to get to know the professors and grad students, and to start to understand how research works.
Read _A PhD is Not Enough_ by Peter Feibelman.
The cost of applying to a grad school is the application fee plus the time it takes to fill out the forms, have GREs and transcripts sent, get letters of recommendation, etc. Apply to as many schools as possible, given these costs. I know a little bit about how admission committees work, and there is a fair amount of randomness - hedge your bets.
Think about why you want to go to grad school. Remember that after a few years of grad school, your stock option loaded ex-undergrad-classmates will be laughing at you.
This is a good time to be applying to grad school in CS. The job market is great, so grad schools are competing for students. Will this still be true in two years? Probably so, but keep an eye out.
After you get accepted at a number of schools, talk to your professors about the schools, and talk on the phone or email professors at the other schools. Visit the ones whose offers you are seriously considering accepting. Go out to lunch with some grad students there, and learn some dirt about the department. It would be a mistake to enter a department without visiting first.
Shamir is touting this design, but not to get people to demand much longer keys. I saw him talk a few weeks ago, and he was careful to emphasize that while the TWINKLE device could make it reasonably easy to crack 512 bit RSA, it won't touch 768 or 1024 bit keys.
The TWINKLE device simply makes factoring large composites of primes a couple of orders of magnitude faster than it is now. The best known factoring algorithms are super-polynomial, so making keys large enough to overcome any constant increase in computing speed is not difficult.
Also, note that the design is not very "blue sky". It is not a general-purpose optical computer. It uses a property of light - that it can be used to implement very large, imprecise adders - to massively speed up part of a factoring algorithm.
Slashdot Mirror
From the hard-to-find Space Child's Mother Goose by Frederick Winsor. This is a hilarious collection of nursery rhymes for scientists.
They're made by Cliff Stoll, of _Cuckoo's Egg_ fame. This book is mandatory reading for hackers.
But, these all refer to classes of computable problems.
The halting problem is the prototypical "not computable" problem. Thus, it is not in NP-hard
Nope. NP-hard and NP-complete refer to decision problems.
The halting problem is a decision problem, and it's at least as hard as satisfiability. Therefore, it is NP-hard.
The classes *I* took used "NP-complete", "NP-hard" and "hard for NP" synonymously.
NP-complete is a subset of NP-hard.
NP-complete problems are all as hard as each other in the sense that either all of them or none of them can be solved by a polynomial-time
algorithm.
NP-hard problems are at least as hard as NP-complete problems. So, for example, the halting problem is NP-hard, but it is not NP-complete.
Anyway, this is not the part of NP-completeness theory that I think programmers need to know. What they need to be able to do is recognize that, for example, the way they are attempting to optimally put outgoing mail messages into the proper queue is equivalent to bin-packing, and therefore takes exponential time in the worst case. They can then go find one of the excellent bin-packing heuristics that has been developed, implement it, and then move on.
A big part of this is simply wrong. Problems are NP-complete if it is proven that there is no algorithm which solves it in polynomial time.
If you can prove the nonexistance of a polynomial time algorithm for any NP-complete problem, you'll get a Turing Award.
That's not the problem. It's not academics, really. Rather, the fault lies with those zealots who claim that anything running a Linux kernel Linux, as though that were all that mattered
Yes - sometimes the problem is ignorance, and sometimes it's zealotry. You might say that a zealot is just the kind of person who takes any difference ("there's only one Linux kernel") and thinks that it's an argument in favor of his favorite system.
Thanks for clarifying,
John Regehr
If those steps could happen, then perhaps progress can be made.
Absolutely. It is important to recognize the inherent tension between vendors trying to differentiate themselves and vendors avoiding scaring off application developers due to the difficulty of targeting (in effect) multiple platforms. This conflict creates a tough problem.
I don't think that they will be, however, because too many people have too much ego wrapped up in the myth.
I don't follow you here. Which academics "repeat their myth about linux=o/s=kernel" and why do they do it? I'm not disagreeing with you - I just have not heard many people making this claim.
John Regehr
The most recent issue of Jakob Nielsen's excellent Alertbox column about web usability is about reputation management. It even mentions slashdot!
Not quite...
/. posts would be calculated only on the basis of my past agreement/disagreement with posters and moderators.
I was proposing that my value function for
If the server compares this profile with other people's profiles, and uses these to augment my profile, then it's collaborative filtering.
Right?
Currently, everybody sees posts that are ranked on the same scale: what the moderators think. I don't agree that this is (always) the right thing to do.
/. based on their preferences.
/. (and the linux kernel dev list, and comp.lang.perl.misc) there are some people whose posts I consistently value. I would like a way to always read their posts. Conversely, I would like a /. "killfile" so that I don't have to read posts by people who have irritated me in the past. Again, I'm not talking about reducing their karma so they can't post - I want a filter so I don't see their posts.
Rather, each reader should have the ability to see a customized view of
For example, if I see a post moderated up to "5" and I think's it's a bonehead post, I would like to de-value the opinions of the moderators who liked that post. This is different from meta-moderation: I'm not saying that these moderators are bad and should be prevented from moderating, I'm saying that I don't want to see the results of their moderations.
Similarly, when I read
Eventually, each user will build up a profile of posters and moderators whose opions they trust and distrust. One can picture people trading these profiles, merging profiles with other people's, etc.
As the amount of information on the net grows, developing a systematic way to figure out which of it you trust and which of it you don't is going to become a very important problem.
Didn't Brin have some sort of reputation management system in Earth? Reading up on this might be a good place to start.
Anyway, the point I want to make is the people have different tastes and want to read different subsets of all posts. There's no reason to force the same view of moderation on everyone.
Tim says:
When you look at all these factors, a price increase of $10 for hardcover over paperback is fairly typical. I'd imagine that you could get by with a $5 spread, but you'd be taking extra risk for benefits that were passed along entirely to the consumer.
An extra $10 for hardcover certainly seems reasonable when we're talking about a $40 softcover, especially since the online bookstores give a slightly larger discount on the hardcover.
So how would O'Reilly go about deciding which books (if any) are worth releasing in hardcover? It's not entirely clear to me that you could make a reliable prediction about something like this through a survey. Maybe it would be obvious just by looking at the 5 or 10 best sellers.
I'd like to see this validated or debunked.
I found this:
Publishing A book which sells for $40 can be produced at a marginal cost of $2. This gap between price and marginal cost has led to a variety of forms of differential pricing. Book clubs, hardcover and paperback editions, and remaindered books are all examples of the ways that the product characteristics and adjusted to support differential pricing.
here. That doesn't make it true, but it sounds reasonable.
"Price discrimination" and "differential pricing" seem to be the keywords.
I was thinking that O'Reilly could attempt to build (even more) customer loyalty by not trying to use price discrimination.
Tim says:
... Despite polls that said people would pay more for a more durable hardcover, everyone bought the softcover to save the difference in price.) So, if there is a poll, how much would you pay for a more durable book? ...
I've heard that the hardcover and softcover versions of a book cost about the same to produce, and that the purpose of a hardcover edition is to extract extra money from people willing to pay more to read a book earlier (since the hardcover comes out well in advance of the softcover).
If this is true, how about releasing hardcover and softcover editions of O'Reilly books at around the same time for around the same price? Seems like it should increase sales and customer satisfaction. I'm definitely in need of a hardcover "Programming Perl"...
As another poster said, you will want to go to a school that has a group that is doing interesting work in an area that you're excited about. This is true, but there are a lot of other factors.
It is important to not only be excited about the reserach, but also to get along with the professor(s) who run the group. Many professors are not good managers - it's important to realize this early on. A big department can be an advantage since there's a better chance that there are other profs doing research that you're interested in if you decide that you can't work with the advisor with whom you originally wanted to work.
If you have two years of undergrad study left, there is a lot you can do to help figure out what you want to do in grad school, and to increase your chances of getting in. First of all, try to take a couple of graduate level CS classes in areas that you're interested in. See if you like them, and if you're comfortable with the workload. Second, you should definitely try to work for a professor in your current department. There are two ways that undergrads can be compensated for their work: money and credit hours. You will end up doing grunt work for a research group, but it's a great way to get to know the professors and grad students, and to start to understand how research works.
Read _A PhD is Not Enough_ by Peter Feibelman.
The cost of applying to a grad school is the application fee plus the time it takes to fill out the forms, have GREs and transcripts sent, get letters of recommendation, etc. Apply to as many schools as possible, given these costs. I know a little bit about how admission committees work, and there is a fair amount of randomness - hedge your bets.
Think about why you want to go to grad school. Remember that after a few years of grad school, your stock option loaded ex-undergrad-classmates will be laughing at you.
This is a good time to be applying to grad school in CS. The job market is great, so grad schools are competing for students. Will this still be true in two years? Probably so, but keep an eye out.
After you get accepted at a number of schools, talk to your professors about the schools, and talk on the phone or email professors at the other schools. Visit the ones whose offers you are seriously considering accepting. Go out to lunch with some grad students there, and learn some dirt about the department. It would be a mistake to enter a department without visiting first.
Be sure to read Olin Shivers's advice to graduate students. Also check out my reading list for computer scientists.
Shamir is touting this design, but not to get people to demand much longer keys. I saw him talk a few weeks ago, and he was careful to emphasize that while the TWINKLE device could make it reasonably easy to crack 512 bit RSA, it won't touch 768 or 1024 bit keys.
The TWINKLE device simply makes factoring large composites of primes a couple of orders of magnitude faster than it is now. The best known factoring algorithms are super-polynomial, so making keys large enough to overcome any constant increase in computing speed is not difficult.
Also, note that the design is not very "blue sky". It is not a general-purpose optical computer. It uses a property of light - that it can be used to implement very large, imprecise adders - to massively speed up part of a factoring algorithm.