As a progressive, I at first assumed that this was regressive Kansas supporting the taxi industry. Until I read the bill. Now I am really reluctant to get in an Uber-dispatched car anywhere, knowing they think a background check, $1M of coverage, and no booze or drugs is too onerous.
Mail order for $800.00 from Concept II [concept2.com]
Rowing is low-impact, aerobic, and you can start as slowly as you like. 30 mins a day while you listen to the radio, watch TV, or just ponder your latest bug.
The unit I mentioned above is suitable for beginners through elite athletes.
Definite nerd appeal with a USB connection and a wireless heart monitor. Lots of builtin stats and uses a plug-in memory card. Regenerative power means a D-cell lasts years.
I'm on my 2nd rowing machine (the first was a competitor but it did last a dozen years and thousands of kms). I'm about to hit 1000 km on this one.
No other $800 piece of exercise equipment will dissipate enough energy (without self-destructing) to give you a decent workout. You'd have to drop more than $3K to get a treadmill anywhere neare as durable. And getting on your feet to walk/run requires a lot more motivation than sitting down on the rower.
You are right. In a 24 hour period since I posted by endorsement, Cogeco cut off my service, accusing me of having a virus and refusing to reinstate me until I had my wife call them back and pretend to be an airhead, then in a separate message accusing me of of violating copyright.
I don't know what has gotten into them, but it is not good.
If you don't like What you got Why don't you change it If your world is all screwed up Rearrange it
Raise a little Hell...
If you don't like what you see Why don't you fight it If you know there's something wrong Why don't you right it
Raise a little Hell...
In the end it comes down to your thinking And there's really nobody to blame When it feels like your ship is sinking And you're too tired to play the game
Nobody's going to help you You've just got to stand up alone And dig in your heels And see how it feels To raise a little Hell of your own
Raise a little Hell...
If you don't like What you got Why don't you change it If your world is all screwed up Rearrange it
I'm perfectly happy with my Cogeco service. I transfer ISOs all the time and get transfer rate in the 5Mbit range. I rarely resort to torrent -- it is unnecessary -- but it seems to work well enough as well. And if there's bandwidth contention, I would want it to be the torrents and not the direct transfers that were hit. That's the point of background.
One the other hand, I have experienced many service providers -- most recently and the Fairmont Banff Springs Hotel -- whose service is crippled.
You are making the specious assumption that students are able to make the connection between paying attention to lectures, doing assignments etc. and their final grades. Many have -- through high school and lame courses -- developed a sense of entitlement that they should be able to do OK while ignoring these educational opportunities and merely skimming the Power Point lecture slides the night before the exam. When they fail many become indignant rather than learning to avail themselves of the opportunity to learn from lectures and assignments.
TopCoder is far more of a race than ACM. ACM give 3 people 5 hours and is (at least at the finals) rarely determined by typing speed or book moves. I'm sure ICFP identifies exceptional candidates just as well.
One caveat: there also exist exceptional candidates who abhor contests. One way to attract them is to do something interesting and see who comes up and says, "cool, how can I get involved?" and opposed to "I am great, how much will you pay me?"
But if you want to hire these superstars, be prepared to compete with the likes of Google, who do both of the above. They have hired better than 1/2 of the programming contest superstars that I have met.
Let me anticipate a strawman response: the tasks posed in programming contests do not represent the sort of tasks that grunt industrial programmers are called on to do. They do select winners.
OK, I'll bite on the off-chance that this isn't a joke. Where does the energy in the battery come from and why is turning it into hydrogen better than, say, leaving it in the battery?
McDonalds isn't "competition" to the soup kitchen any more than the Asus Eee is competition to the XO Laptop.
You'd be wrong. Intel-insides are actively bidding against OLPC in developing markets. Whether or not the bids are worthy, they are backed by the normal corporate dirty tricks -- including FUD and dumping to name two -- and aimed to kill.
Why is Intel's departure a blow? Why is a non-competition agreement such an unreasonable thing to expect of a partner? I daresay OLPC's take (which has not yet been stated in the media) is that Intel was helping themselves to inside information and offering little in return.
It would have been nice if Intel and OLPC could have come up with an arrangement to differentiate themselves in the developing world market, but it didn't work out. So they go it alone. The computers are quite different, the OLPC being designed from the ground up for its purpose, the Classmate and friends being crippled conventional laptops.
And whether or not Intel and friends manage to kill OLPC, they wouldn't have had a dog in the race at all if not for OLPC.
Rogers are clearly not inserting content into users' web pages, as the title claims. They are inserting content into pages viewed by users.
So I have little faith in the claim that they are "intercepting http." What is more likely is that the default proxy server they provide is inserting the content. While it may make little difference to the average user, as the "normal" setup uses the proxy, it seems to me that there's a huge difference between supplying a proxy and intercepting and manipulating http traffic; that is, hijacking TCP port 80. The proxy I can easily avoid by using a direct connection to the internet; TCP hijacking, I can't.
"And I didn't say either. I said that "over 99% accuracy" (the claim I saw in TFA) is not that impressive anymore, and it's not. My Bayesian filter has been doing better than that for years."
What performance guarantee do or did you give your clients? That's what the 99% is. A performance guarantee. For all comers. No training required. Would that be a yawn for you?
You require some training, your results are for one user only -- motivated and dedicated one at that, and we have your impressions to quantify the degree of reporting.
If you want to correspond more, email me. I'm sure/. readers are bored to tears.
> You seem to be interested in super-exact statistics relating to spam filtering. I do not claim to have investigated it with that level of precision.
I am interested in *accurate* statistics. You are the one who used 4 significant figures to quote an accuracy figure 99.77% in support of your argument that your Bayesian filter was superior to a method whose accuracy was reported with much less precision: "at least 99%." If you'd said "about 99 1/2 percent" I probably wouldn't have jumped on you, but your statement wouldn't have had as much impact, would it?
> What I can tell you is that my Bayesian filter works with a high enough success against spam and with a low enough false positive rate that, to me, spam isn't a problem for me.
I call that "faith based evaluation."
> I still avoid posting my email address on the web but I no longer bother using temporary email addresses when buying something online, etc.
I do neither of these avoidance techniques. They compromise my ability to communicate.
Your filter is good enough for you and your clients. Fine. But when you say it is as good as it can get, or that it is better than something else, be prepared to justify your claims.
> I've used the same email address since 1993. I get 30,000+ messages per month, 98.6% of which is spam. I have spam statistics on a monthly basis back to November 2002. In total, I've received 904,202 spams since May 27th, 2004--99.83% of which were caught by the Bayesian filter during that time. Last month, 99.91% of my spam was caught.
So you get 30,000 spams a month, and 1.4%*30,000 = 420 legit emails/month.
> I'm not on a crusade to prove accuracy so, no, I didn't do it twice. These are the real-world results I'm getting. Is it possible I have missed a false positive or two? Over the last 4 years, sure.
48 months * 420 legit/month = 20,160 emails.
If it is really "a false positive or two" that's a 1/10,000 error rate. But if it's really "a false positive or two per year" it's 1/2500. And if it's really "a false positive or two per month" it's 1/210. Unless you searched through the 30,000 spams you received in that month, how would you know?
> If this response doesn't answer your questions, let me know and I'll email you.
Contact me anyway. I'm always interested in evaluation methodology and experiments that demonstrate the efficacy of spam filters. Maybe we can collaborate. Have you run your filter on the TREC corpora? They simulate exactly the sort of deployment you're talking about. Under laboratory conditions the best filters get the sort of results you're talking about, but transferability to the field has yet to be established. And the best filters aren't what I'd call "Bayesian."
>> 99.77% sounds a bit high for your Bayesian filter
> Yet that's the success rate I'm getting.
We should probably take this off-line. Drop me a line at gvcormac@uwaterloo.ca (Email address obfuscation is a concession of defeat to the spammers.)
99.77% is not preposterous, but I would be interested to know the methods you used to measure this, and in particular how you know that errors are not underreported. To measure 99.77% accuracy you'd need tens of thousands of messages. Did you really read them all carefully and adjudicate them without any access to (and hence influence from) the filter's opinion? Did you do it twice so as to measure your own reliability? Did you provide totally accurate and immediate feedback to your filter in real-time, whenever it made a mistake?
> Bayesian requires that you initially train it. Once it's reasonably trained, it essentially > trains itself. Yes, I report the occasional spam that gets through. But with 99.77% > accuracy, that's not very often.
This is true, but "not much training" is materially different from "no training." Personally, I find training a personal filter not too onerous, but I'm atypical, and so are you.
99.77% sounds a bit high for your Bayesian filter -- I bet you are missing some errors and/or biasing your judgements to conform to those of your filter and/or not counting "gray mail" which could go either way.
However, for the sake of argument, I'll accept your numbers. The major difference between your filter and the one in TFA is that yours requires that you continually train it and the one in TFA doesn't. A 10% error rate is not uncommon for server-side filters which don't accept user training. This method competes with them, and it typically does give numbers closer to 99.9% than the 99% that the vendor guarantees.
"Spam detection has got to be something like 99.999% accurate"
Nonsense. 99.999% is one error in 100,000 emails. Have you even received 100,000 emails? Have you checked every one to see if the filter made at most one mistake? Have you repeated the measurement several dozen times, as would be necessary to make such a claim? Of course not.
I would be surprised if the filter you are using (including Gmail) is 99% accurate.
Here are some accuracy figures under ideal conditions. From side-by-side comparisons I can assure you that spam filters in the field do considerably worse. You just don't notice.
Canadian Tire is a franchise operation. While the corporation has, I'm sure, overall guidelines and standards, the quality of service you get will depend a fair amount on how the franchise is run.
Personally, I find that their parts are 1/2 the price of the competition and just as good, and the quality of work has about the same mean and variance as elsewhere. I like the service manager at my local franchise and any time I've had problems with the work they have fixed it with no hassle and no charge.
That said, I prefer to fix my car and my computer myself. It is less hassle, cheaper, and usually quicker.
Slashdot Mirror
As a progressive, I at first assumed that this was regressive Kansas supporting the taxi industry. Until I read the bill. Now I am really reluctant to get in an Uber-dispatched car anywhere, knowing they think a background check, $1M of coverage, and no booze or drugs is too onerous.
Perhaps I am missing something?
[from http://ask.slashdot.org/comments.pl?sid=141278&cid=11855885 ]
Mail order for $800.00 from Concept II [concept2.com]
Rowing is low-impact, aerobic, and you can start
as slowly as you like. 30 mins a day while you
listen to the radio, watch TV, or just ponder your
latest bug.
The unit I mentioned above is suitable for
beginners through elite athletes.
Definite nerd appeal with a USB connection and
a wireless heart monitor. Lots of builtin
stats and uses a plug-in memory card.
Regenerative power means a D-cell lasts years.
I'm on my 2nd rowing machine (the first was
a competitor but it did last a dozen years
and thousands of kms). I'm about to hit 1000
km on this one.
No other $800 piece of exercise equipment will
dissipate enough energy (without self-destructing)
to give you a decent workout. You'd have to
drop more than $3K to get a treadmill anywhere
neare as durable. And getting on your feet to
walk/run requires a lot more motivation than
sitting down on the rower.
You are right. In a 24 hour period since I posted by endorsement, Cogeco cut off my service, accusing me of having a virus and refusing to reinstate me until I had my wife call them back and pretend to be an airhead, then in a separate message accusing me of of violating copyright.
I don't know what has gotten into them, but it is not good.
Bell is worse, so what am I to do?
If you don't like
...
...
...
...
What you got
Why don't you change it
If your world is all screwed up
Rearrange it
Raise a little Hell
If you don't like what you see
Why don't you fight it
If you know there's something wrong
Why don't you right it
Raise a little Hell
In the end it comes down to your thinking
And there's really nobody to blame
When it feels like your ship is sinking
And you're too tired to play the game
Nobody's going to help you
You've just got to stand up alone
And dig in your heels
And see how it feels
To raise a little Hell of your own
Raise a little Hell
If you don't like
What you got
Why don't you change it
If your world is all screwed up
Rearrange it
Raise a little Hell
http://www.trooper.ca/default.php?cat=lyrics&subcat=56
I'm perfectly happy with my Cogeco service. I transfer ISOs all the time and get transfer rate in the 5Mbit range. I rarely resort to torrent -- it is unnecessary -- but it seems to work well enough as well. And if there's bandwidth contention, I would want it to be the torrents and not the direct transfers that were hit. That's the point of background.
One the other hand, I have experienced many service providers -- most recently and the Fairmont Banff Springs Hotel -- whose service is crippled.
You are making the specious assumption that students are able to make the connection between paying attention to lectures, doing assignments etc. and their final grades. Many have -- through high school and lame courses -- developed a sense of entitlement that they should be able to do OK while ignoring these educational opportunities and merely skimming the Power Point lecture slides the night before the exam. When they fail many become indignant rather than learning to avail themselves of the opportunity to learn from lectures and assignments.
There's no free lunch. The electrolysis takes exactly as much energy as you get out of burning the hydrogen. Plus overhead.
TopCoder is far more of a race than ACM. ACM give 3 people 5 hours and is (at least at the finals) rarely determined by typing speed or book moves. I'm sure ICFP identifies exceptional candidates just as well.
One caveat: there also exist exceptional candidates who abhor contests. One way to attract them is to do something interesting and see who comes up and says, "cool, how can I get involved?" and opposed to "I am great, how much will you pay me?"
But if you want to hire these superstars, be prepared to compete with the likes of Google, who do both of the above. They have hired better than 1/2 of the programming contest superstars that I have met.
Let me anticipate a strawman response: the tasks posed in programming contests do not represent the sort of tasks that grunt industrial programmers are called on to do. They do select winners.
OK, I'll bite on the off-chance that this isn't a joke. Where does the energy in the battery come from and why is turning it into hydrogen better than, say, leaving it in the battery?
Hydrogen is a method for transmission and storage of energy. It is not a source of energy. At least not until they figure out controlled fusion.
McDonalds isn't "competition" to the soup kitchen any more than the Asus Eee is competition to the XO Laptop.
You'd be wrong. Intel-insides are actively bidding against OLPC in developing markets. Whether or not the bids are worthy, they are backed by the normal corporate dirty tricks -- including FUD and dumping to name two -- and aimed to kill.
Why is Intel's departure a blow? Why is a non-competition agreement such an unreasonable thing to expect of a partner? I daresay OLPC's take (which has not yet been stated in the media) is that Intel was helping themselves to inside information and offering little in return.
It would have been nice if Intel and OLPC could have come up with an arrangement to differentiate themselves in the developing world market, but it didn't work out. So they go it alone. The computers are quite different, the OLPC being designed from the ground up for its purpose, the Classmate and friends being crippled conventional laptops.
And whether or not Intel and friends manage to kill OLPC, they wouldn't have had a dog in the race at all if not for OLPC.
Rogers are clearly not inserting content into users' web pages, as the title claims. They are inserting content into pages viewed by users.
So I have little faith in the claim that they are "intercepting http." What is more likely is that the default proxy server they provide is inserting the content. While it may make little difference to the average user, as the "normal" setup uses the proxy, it seems to me that there's a huge difference between supplying a proxy and intercepting and manipulating http traffic; that is, hijacking TCP port 80. The proxy I can easily avoid by using a direct connection to the internet; TCP hijacking, I can't.
"And I didn't say either. I said that "over 99% accuracy" (the claim I saw in TFA) is not that impressive anymore, and it's not. My Bayesian filter has been doing better than that for years."
/. readers are bored to tears.
What performance guarantee do or did you give your clients? That's what the 99% is. A performance guarantee. For all comers. No training required. Would that be a yawn for you?
You require some training, your results are for one user only -- motivated and dedicated one at that, and we have your impressions to quantify the degree of reporting.
If you want to correspond more, email me. I'm sure
> You seem to be interested in super-exact statistics relating to spam filtering. I do not claim to have investigated it with that level of precision.
I am interested in *accurate* statistics. You are the one who used 4 significant figures to quote an accuracy figure 99.77% in support of your argument that your Bayesian filter was superior to a method whose accuracy was reported with much less precision: "at least 99%." If you'd said "about 99 1/2 percent" I probably wouldn't have jumped on you, but your statement wouldn't have had as much impact, would it?
> What I can tell you is that my Bayesian filter works with a high enough success against spam and with a low enough false positive rate that, to me, spam isn't a problem for me.
I call that "faith based evaluation."
> I still avoid posting my email address on the web but I no longer bother using temporary email addresses when buying something online, etc.
I do neither of these avoidance techniques. They compromise my ability to communicate.
Your filter is good enough for you and your clients. Fine. But when you say it is as good as it can get, or that it is better than something else, be prepared to justify your claims.
> I've used the same email address since 1993. I get 30,000+ messages per month, 98.6% of which is spam. I have spam statistics on a monthly basis back to November 2002. In total, I've received 904,202 spams since May 27th, 2004--99.83% of which were caught by the Bayesian filter during that time. Last month, 99.91% of my spam was caught.
So you get 30,000 spams a month, and 1.4%*30,000 = 420 legit emails/month.
> I'm not on a crusade to prove accuracy so, no, I didn't do it twice. These are the real-world results I'm getting. Is it possible I have missed a false positive or two? Over the last 4 years, sure.
48 months * 420 legit/month = 20,160 emails.
If it is really "a false positive or two" that's a 1/10,000 error rate. But if it's really "a false positive or two per year" it's 1/2500. And if it's really "a false positive or two per month" it's 1/210. Unless you searched through the 30,000 spams you received in that month, how would you know?
P.S. Do I know you?
> If this response doesn't answer your questions, let me know and I'll email you.
Contact me anyway. I'm always interested in evaluation methodology and experiments that demonstrate the efficacy of spam filters. Maybe we can collaborate. Have you run your filter on the TREC corpora? They simulate exactly the sort of deployment you're talking about. Under laboratory conditions the best filters get the sort of results you're talking about, but transferability to the field has yet to be established. And the best filters aren't what I'd call "Bayesian."
Here are the results to beat: http://www.eecs.tufts.edu/~dsculley/papers/emailAndWebSpamSIGIR.pdf
>> 99.77% sounds a bit high for your Bayesian filter
> Yet that's the success rate I'm getting.
We should probably take this off-line. Drop me a line at gvcormac@uwaterloo.ca (Email address obfuscation is a concession of defeat to the spammers.)
99.77% is not preposterous, but I would be interested to know the methods you used to measure this, and in particular how you know that errors are not underreported. To measure 99.77% accuracy you'd need tens of thousands of messages. Did you really read them all carefully and adjudicate them without any access to (and hence influence from) the filter's opinion? Did you do it twice so as to measure your own reliability? Did you provide totally accurate and immediate feedback to your filter in real-time, whenever it made a mistake?
> Bayesian requires that you initially train it. Once it's reasonably trained, it essentially > trains itself. Yes, I report the occasional spam that gets through. But with 99.77%
> accuracy, that's not very often.
This is true, but "not much training" is materially different from "no training." Personally, I find training a personal filter not too onerous, but I'm atypical, and so are you.
99.77% sounds a bit high for your Bayesian filter -- I bet you are missing some errors and/or biasing your judgements to conform to those of your filter and/or not counting "gray mail" which could go either way.
However, for the sake of argument, I'll accept your numbers. The major difference between your filter and the one in TFA is that yours requires that you continually train it and the one in TFA doesn't. A 10% error rate is not uncommon for server-side filters which don't accept user training. This method competes with them, and it typically does give numbers closer to 99.9% than the 99% that the vendor guarantees.
"Spam detection has got to be something like 99.999% accurate"
Nonsense. 99.999% is one error in 100,000 emails. Have you even received 100,000 emails? Have you checked every one to see if the filter made at most one mistake? Have you repeated the measurement several dozen times, as would be necessary to make such a claim? Of course not.
I would be surprised if the filter you are using (including Gmail) is 99% accurate.
Here are some accuracy figures under ideal conditions. From side-by-side comparisons I can assure you that spam filters in the field do considerably worse. You just don't notice.
22425 spams in my quarantine. 747.83333 spams/day.
A mere trickle.
Canadian Tire is a franchise operation. While the corporation has, I'm sure, overall guidelines and standards, the quality of service you get will depend a fair amount on how the franchise is run.
Personally, I find that their parts are 1/2 the price of the competition and just as good, and the quality of work has about the same mean and variance as elsewhere. I like the service manager at my local franchise and any time I've had problems with the work they have fixed it with no hassle and no charge.
That said, I prefer to fix my car and my computer myself. It is less hassle, cheaper, and usually quicker.
Sorry, mangled the URL in the parent: Wumpus-Search.org
It is free, libre. Wumpus Search.