The thing you do isn't to create just a binary dichotomy of good vs. bad, but to actually put things *into* categories for Nazi or gambling. The good/bad determination is then done afterward by taking the output of the Bayesian analysis and saying that anything that the Bayesian tool says is a gambling site goes into the "bad" category.
It's just that most current implemenations of Bayesian filtering are focused on a binary problem: spam. Even there, though, software like POPFile does quite well keeping my work and personal stuff seperate from each other in addition to seperating spam.
I've run well over 100,000 adult sites through my training program on different occasions. The list is easily obtained in XML format from the DMOZ.org project.
I've got one about half done that works OK and have been messing with Bayesian analysis for both filtering of content and recommendation of content in both web sites and RSS feeds. I'm currently using a PHPBayesian class and using an iframed script as my "proxied browser" because I am comfortable with PHP and it makes for a decent testbed. That system just spits out large text files for use in software like Squid. If you wanted more ongoing, real-time evaluation, there'd need to be an implementation sitting at the proxy level to do evalutation on each URL that isn't already in the blacklist on the fly, which would require a more speedy implementation than my current testbed.
I intend on making the filtering engine open source and possibly build a Mozilla toolbar for select people to use and do the training correction with their browsing to replace my iframed script.
Ideally, different data sets could be used to build different filters: one for elementary schools, one for locked down workplaces, one for less locked down workplaces, etc.
The nice thing about this is that the categories don't need to be just porn, you can also include cracking, gambling, etc. easily. Of course the software could also be used for politically slanted lists as well, but any flexible tool can be used in inappropriate ways. I wouldn't necessarily agree with it, but it could easily be used to build a church-defined filter with very specific criteria to shield parishners from whatever evil their theology objects to.
That's part of why I've got a computer sitting at home today churning through sites trying to train a Bayesian engine to detect inappropriate stuff. Once it's done, putting it in between the proxy's lookup and handing over of the page will allow learning and building a bigger list. The first engine will be used to build a filter intended for children under 13. Future engines will analyze for different criteria. I intend to charge for the list subscriptions, but will likely make the generating software open source and will gladly share the criteria/databases.
I'm currently using self-classified content/sites from DMOZ.org. If a site has voluntarily added itself to the Adult section on DMOZ, they've already indicated that they are not appropriate for children. This also means that when I get to the filter for older children (initial target fo the lists is schools), I can explicitly allow those sites categorized as educational rather than outright porn and get different lists. Since DMOZ is human reviewed and sites are categorized by experts, that's the data set I'm using to train my filter.
It does OK right now, but I'm looking for better methods for training it to approach that 99.9% effective rate that I'm currently getting with POPFile on my spam.
4 years ago I signed up with MediaOne cable, which became RoadRunner cable, which became AT&T broadband, which finally became Comcast. Along the way, each left traces that took *years* to finally transition all of the way.
This is also why I use Subversion (used to use CVS) repositories even when I'm the only developer on a project. Regular commit cycles give you a dated trail for the development of code and you can show the evolution of code rather than the sudden arrival of entire functions and classes the are fully functional.
Actually, a surprising number of those FOSS Linux apps work on Windows as well. With a combination of Cywin and native ports, I run most of my favorite Linux utilities on my Windows workstations.
Yeah. Because you had a chance. Dear Lord, if your fantasy life can't cope with marital status in your fantasy targets, you may need a little more imagination.
When you consider how much TV most people watch, if you dropped the satellite/cable and went with 8 at a time Netflix, you'd actually either have to drop the amount you watch or run out even at 8 at a time.
I live in Minnesota and love it. The cold keeps the riff raff out. See, you talk like sun and heat is a good thing and that is nearly impossible for me to understand. I love the cold and find a stiff October/November breeze to be about the best weather possible. I suffer through the 2.5 months of 80°F that we *do* get and live for the cooler weather. I don't bother with a jacket until it gets down around freezing.
You know that warm happy feeling you get in the sun? Some of us get that exact same feeling when its 36°F with a stiff north wind and we're outside to enjoy it. It's in our blood and our ancestors went where they could get it.
I don't think I took a single book home between my sophomore and senior year in high school and stayed on the honor roll.
Doing well in high school is not dependant on doing any homework. I took a lot of classes where things were set up that doing well on the tests could garner a decent overall grade. I always made sure I had a study hall and actually used it to complete the little homework I did do.
I can't believe I'm going to be defending vegetarians, but not everyone who calls themselves vegetarian does so for idealogical reasons. There are some who choose to do so for flavor reasons (much the same reason I *can't* be a vegetarian), financial reasons (per pound, vegetables and fruit are much cheaper than meat), etc. That doesn't make them fake vegetarians, just not idealogical ones.
Actually, this is a problem for some people (including me) with even more expensive pedometers. A couple of months ago, I bought a $20 one and watched it either way overcount (3:1) or way undercount (5 minute walk that resulted in 100 steps registering). I'm reluctant to shell out for an expensive one given this track record.
That stops 3rd generation copies, but the article talks about preventing >n copies in the *2nd* generation where n is the maximum number of allowed copies of a given disc. Without 3rd party tracking or modifying the disc, how do you track n?
Actually Slashdot is a place where the horse has been flogged until all that remains to be flogged is the memory of the dead horse and even that is wearing out.
I use the StumbleUpon toolbar for my web "channel surfing". It gives me a lot of stuff I've already seen, but also quite a bit of new/interesting stuff as well.
Re:I saw this last night, some interesting points.
on
The Way the Music Died
·
· Score: 1
They didn't talk to Petty? "The Last DJ" is the most direct anti-music industry music I've heard in a long time if not ever before. I love that album (of course I own every one of his so I may be biased). The title song's main character is painted as a huge rebel and calls him the last DJ who plays what he wants to play. It goes on like that song after song, detailing how money has ruined music. Particularly listen to the track called "Joe" which is about a music company CEO who asks for an "angel whore who can learn a guitar lick" to make more money.
In short, if you like Tom Petty at all and aren't happy with the current state of the music industry, it's definitely an album worth buying.
Slashdot Mirror
It might be that the processors that they *do* buy will be higher end ones instead of the cheap ones.
The thing you do isn't to create just a binary dichotomy of good vs. bad, but to actually put things *into* categories for Nazi or gambling. The good/bad determination is then done afterward by taking the output of the Bayesian analysis and saying that anything that the Bayesian tool says is a gambling site goes into the "bad" category.
It's just that most current implemenations of Bayesian filtering are focused on a binary problem: spam. Even there, though, software like POPFile does quite well keeping my work and personal stuff seperate from each other in addition to seperating spam.
I've run well over 100,000 adult sites through my training program on different occasions. The list is easily obtained in XML format from the DMOZ.org project.
I've got one about half done that works OK and have been messing with Bayesian analysis for both filtering of content and recommendation of content in both web sites and RSS feeds. I'm currently using a PHPBayesian class and using an iframed script as my "proxied browser" because I am comfortable with PHP and it makes for a decent testbed. That system just spits out large text files for use in software like Squid. If you wanted more ongoing, real-time evaluation, there'd need to be an implementation sitting at the proxy level to do evalutation on each URL that isn't already in the blacklist on the fly, which would require a more speedy implementation than my current testbed.
I intend on making the filtering engine open source and possibly build a Mozilla toolbar for select people to use and do the training correction with their browsing to replace my iframed script.
Ideally, different data sets could be used to build different filters: one for elementary schools, one for locked down workplaces, one for less locked down workplaces, etc.
The nice thing about this is that the categories don't need to be just porn, you can also include cracking, gambling, etc. easily. Of course the software could also be used for politically slanted lists as well, but any flexible tool can be used in inappropriate ways. I wouldn't necessarily agree with it, but it could easily be used to build a church-defined filter with very specific criteria to shield parishners from whatever evil their theology objects to.
That's part of why I've got a computer sitting at home today churning through sites trying to train a Bayesian engine to detect inappropriate stuff. Once it's done, putting it in between the proxy's lookup and handing over of the page will allow learning and building a bigger list. The first engine will be used to build a filter intended for children under 13. Future engines will analyze for different criteria. I intend to charge for the list subscriptions, but will likely make the generating software open source and will gladly share the criteria/databases.
I'm currently using self-classified content/sites from DMOZ.org. If a site has voluntarily added itself to the Adult section on DMOZ, they've already indicated that they are not appropriate for children. This also means that when I get to the filter for older children (initial target fo the lists is schools), I can explicitly allow those sites categorized as educational rather than outright porn and get different lists. Since DMOZ is human reviewed and sites are categorized by experts, that's the data set I'm using to train my filter.
It does OK right now, but I'm looking for better methods for training it to approach that 99.9% effective rate that I'm currently getting with POPFile on my spam.
No, but a search for "whitehouse" shouldn't bring up whitehouse.com before whitehouse.gov.
4 years ago I signed up with MediaOne cable, which became RoadRunner cable, which became AT&T broadband, which finally became Comcast. Along the way, each left traces that took *years* to finally transition all of the way.
This is also why I use Subversion (used to use CVS) repositories even when I'm the only developer on a project. Regular commit cycles give you a dated trail for the development of code and you can show the evolution of code rather than the sudden arrival of entire functions and classes the are fully functional.
Actually, a surprising number of those FOSS Linux apps work on Windows as well. With a combination of Cywin and native ports, I run most of my favorite Linux utilities on my Windows workstations.
The Ex-Lax is in the cabinet.
Lewis Black puts it this way. "The Democrats are the party of no ideas. The Republicans are the party of bad ideas."
Yeah. Because you had a chance. Dear Lord, if your fantasy life can't cope with marital status in your fantasy targets, you may need a little more imagination.
Or for only $499, you can get an iMow from Toro.
I add to that and say that if you try, you'll find yourself supporting 9 times more baby than you planned for.
I use Nonna Urbixness as the first name and last name.
When you consider how much TV most people watch, if you dropped the satellite/cable and went with 8 at a time Netflix, you'd actually either have to drop the amount you watch or run out even at 8 at a time.
I live in Minnesota and love it. The cold keeps the riff raff out. See, you talk like sun and heat is a good thing and that is nearly impossible for me to understand. I love the cold and find a stiff October/November breeze to be about the best weather possible. I suffer through the 2.5 months of 80°F that we *do* get and live for the cooler weather. I don't bother with a jacket until it gets down around freezing.
You know that warm happy feeling you get in the sun? Some of us get that exact same feeling when its 36°F with a stiff north wind and we're outside to enjoy it. It's in our blood and our ancestors went where they could get it.
I don't think I took a single book home between my sophomore and senior year in high school and stayed on the honor roll.
Doing well in high school is not dependant on doing any homework. I took a lot of classes where things were set up that doing well on the tests could garner a decent overall grade. I always made sure I had a study hall and actually used it to complete the little homework I did do.
I can't believe I'm going to be defending vegetarians, but not everyone who calls themselves vegetarian does so for idealogical reasons. There are some who choose to do so for flavor reasons (much the same reason I *can't* be a vegetarian), financial reasons (per pound, vegetables and fruit are much cheaper than meat), etc. That doesn't make them fake vegetarians, just not idealogical ones.
Actually, this is a problem for some people (including me) with even more expensive pedometers. A couple of months ago, I bought a $20 one and watched it either way overcount (3:1) or way undercount (5 minute walk that resulted in 100 steps registering). I'm reluctant to shell out for an expensive one given this track record.
That stops 3rd generation copies, but the article talks about preventing >n copies in the *2nd* generation where n is the maximum number of allowed copies of a given disc. Without 3rd party tracking or modifying the disc, how do you track n?
"try running over a rebel with a car."
How about not leaving $1000 equipment where someone can run over it?
Actually Slashdot is a place where the horse has been flogged until all that remains to be flogged is the memory of the dead horse and even that is wearing out.
I use the StumbleUpon toolbar for my web "channel surfing". It gives me a lot of stuff I've already seen, but also quite a bit of new/interesting stuff as well.
They didn't talk to Petty? "The Last DJ" is the most direct anti-music industry music I've heard in a long time if not ever before. I love that album (of course I own every one of his so I may be biased). The title song's main character is painted as a huge rebel and calls him the last DJ who plays what he wants to play. It goes on like that song after song, detailing how money has ruined music. Particularly listen to the track called "Joe" which is about a music company CEO who asks for an "angel whore who can learn a guitar lick" to make more money.
In short, if you like Tom Petty at all and aren't happy with the current state of the music industry, it's definitely an album worth buying.