Hacker News new | past | comments | ask | show | jobs | submit login
Genetic A/B Testing with JavaScript (ejohn.org)
46 points by jeresig on Nov 26, 2008 | hide | past | web | favorite | 14 comments



As I said in the other thread, if we somehow (javascript?) could get click data on Google Adsense ads with its parameters (colors, size), we could genetically optimize them, which would be tremendously profitable.


http://www.techcrunch.com/2007/10/19/yieldbuild-will-try-to-...

they seem to do what you're talking about with adsense


Cool nice find! At least it shows that it's possible :) It's free to try for 30 days, i'll give it a spin - after that they ask for 3%, which doesn't seem too bad, but is still quite a bit of money - the question is, will their service bring me more than that? Probably, let's see!

Still hoping for an open-source version though. I find it strange that Google itself doesn't allow this, they would win more money, too.


If you start making a good amount from adsense, I think Google take an interest.

I was contacted by Google and offered an "Optimization" - basically they looked over one of my websites, and moved a few things around, changed a few settings on the adsense, and emailed me back some mockup jpegs. They just said "We think this will work better, it's up to you if you'd like to try it" - I did, and it increased revenue... They mailed back a little later to say it looked like it had a good effect etc.

I know it's not an automatic clever machine based system, but it worked well for me.


Both of these are cool but add a query to their server on each pageview, which slows down a site - and what if their server goes offline for 2 hours? - for this reason, i think a hosted solution would still be best.


Just found Pubmatic which does the same thing, too - http://www.pubmatic.com/

It's free as it's in beta now - can't find how much it will cost in the future.


I love the idea! I think that implementing all the content in HTML and then selecting it in Javascript though is a horrible idea for a number of reasons. Performance and accessibility would be my top two. Maintainability would be a close third.


you're absolutely right but it'd be straightforward to implement the technique server side to fix those problems.


I agree. I think this could make a really interesting project in PHP/Python/Rails/etc


Offtopic : I am frustrated with people ignoring no free lunch theorem and trying to use GAs everywhere.


The NFLT mainly applies across all functions. So, I don't see a prima facie reason that there can't be a best overall algorithm for the problems we tend to encounter, since these are an infinitely small subset of all possible problems and likely share common characteristics.

In fact, to be a total contrarian, I suspect intelligent design theory can provide such a heuristic for all real world problem domains. The irony is thick indeed.


> Offtopic : I am frustrated with people ignoring no free lunch theorem and trying to use GAs everywhere.

GAs don't violate No Free Lunch. After all, you're a genetic algorithm...


That isn't what he's saying. He thinks people should focus more on the problem domain rather than just use GAs indiscriminately.


Then his "offtopic" disclaimer is apt.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: