Feel free delve into any questions.
* Filter results by eBay category.
* Outlier elimination such as excluding "broken" items (e.g. listings containing commonly used words such as "spares or repair", "cracked screen", "damaged").
* Country specific currency (e.g. for UK searches show the results in GBP, £).
* Predict a fair market value for a given search. Useful when trying to sell items second hand as an "arm's-length transaction".
I must believe there's any easy way to eliminate some "outliers" using mathematics, but I can't recall the function(s) to do so.
The median is one good way, as you already have. You can also use the interquartile mean: http://en.wikipedia.org/wiki/Interquartile_mean
Perhaps I need to filter it within 1 or 1.5 stdevs. Will experiment with this.
However, sometimes you can easily see there are two clusters of results. Not sure how to mathematically determine this. Any ideas?
Just pointing out that there is a decent sized market for this type of data.
Locksley, keep up the good work. You need to clean up the results. Removing the unrelated items is desirable. This has a good monetizing potential.
Yeah, I've come across Pricenomics but found their data inconsistent and the sample size too small with ebay results. They're doing good work nonetheless and their front end rapes mine haha. I ain't much of a designer.
This article explains it better than I ever will.
http://blog.codeship.io/2012/05/06/Unicorn-on-Heroku.html
I'm running 8 unicorn workers per dyno at the moment and that's still within the 512mb memory limit because I'm not using ActiveRecord.