New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Hi @tehdan, I've been reviewing some of these networks. It's mainly small ISPs that offer hosting / support for hosting in residential networks. It's not really an issue of EU networks or any region in general. Bigger networks are more accurate because there's more data. In the past few months, some new code was added to greylist certain networks and some post processing to try to make it more accurate. Most clients that want the least amount of false positives uses flags=m option (as noted on the website). When it comes to learning algorithms, false positives can't really be avoided. In general, I stand by the results it generates. Let's say a value of 0.99 is produced by an IP address. For all IPs that produce 0.99, 1/100 is a false positive. Of course, the threshold to take action is based on the person using it. Someone can take action for values > 0.995 or > 0.999. I always recommend people to test their datasets before integration so they can find the correct threshold.
I'd be happy to look at any network / IP address that you believe is wrong. Sometimes, human intervention is required for learning algorithms
So if an home ISP also runs hosting related services in their network you will flag it? That sounds rather problematic, an example would be KPN in the Netherlands its the biggest ISP in this country but it does provide much more than a internet line, would all KPN users end up flagged?
If you use flags=m it shouldn't be a problem, but if you want to use the learning algorithm, then what looks like a hosting network will generate high values. It becomes a trade-off of how many proxy / VPN / bad IPs you want to let through vs how many false positives that occur. Of course, the person implementing their code has full control of what values they want to take action on and what flags that is used. This is why I recommend people test the system first with their datasets or else the value they set to take action on becomes arbitrary. I have some post processing tools which allows me to reduce the result if necessary. Can you try some IPs with the system and see if high values are generated? If you see values > 0.98 please let me know.
I tried a kpn ip and im getting -6 as a result (no flags used) when i use http://check.getipintel.net/check.php?ip=
after i tried the web interface on getipintel.net and this returned:
blacklisting kpn is unacceptable because its truely powering the netherlands, many companies resell their resources.
@Mark_R can you PM me the IP please? -6 error means you didn't provide any contact information, which is required when using the API. If you're looking up an IP address with the web interface and you're on a proxy / VPN, it'll return that error.
pm'd
The IP you PM'd me returned a value of "0"
my worry is that you might override this result because kpn provides more than just home isp services causing anyone that uses your api to auto reject orders from a majority of the netherlands.
my worry is that you might override this result because kpn provides more than just home isp services causing anyone that uses your api to auto reject orders from a majority of the netherlands.
We do not ban ISPs that offer both residential services & hosting. A lot of big ISPs do that so there'd be too many false positives
We utilize a robust integration of deterministic cross-synergies to pollinate the ornated outcome in a planar-multitude scale of infrustructure.
Something like that, yeah. Machine learning is a variation of the boosting technique but with quite a few adjustments. Probability theory techniques are mainly from reliability modeling.
TL;DR - math n unicorns give decent results.