hi,ppl.

the runecrypt graph for prices is it true or false.
i mean when u wright rune scimi and see how much they go up 2morow,
is it working or its fake.
pm me plz

It is very much real. With the runecrypt graph, it generates the prices every 12 hrs unlike Jagex who updates the prices at random times. So sometimes the graphs on runecrypt might be a few hours delayed but its a great guide to use.

I think he's talking about the price predicting.

It's very unlikely to be very accurate. usually you just see a straight like that goes on forever as if the last price change will go on and on and on.

It's not a matter of "true or false" but rather "accurate or inaccurate." Since I've never used it I can't say whether it is accurate or not, but- don't overly rely on it.

add me on runescape perper88, for merching.

are you part of a clan kungfu or are you just looking about the basics of merchanting?

The prediction line follows the average trend of the items price, and displays the estimated price accordingly. Prices may temporarily flucuate up or down, but unless something drastic happens, the price should eventually (within a week or two) level off back to the estimated line.

PS: We're not liable if something bad happens, as a result of our predictions =@

Hmmm... might be worth throwing in an official asterisk and disclaimer on the graph page to notify users not to rely on the predictions too heavily... The algorithm is very basic (just a linear trend line) so don't read it as though the prices are set in stone for the future.

I wonder if it could be possible to have not just a linear regression, but also quadratic, cubic, quart.. regressions? After all, if a TI83 can do it easily, it shouldn't be that hard to have a computer program do it right?

It wouldn't be too bad, but i'd have to crank out my own formulas for those regressions and my math skills are "rusty" to say the least. I'll work on it, but if I do add those I'll probably put them in an "advanced options" box or something because it could be potentially very confusing for some users. In the future I would like to buckle down and really dive into the math, and develop a means of choosing the best type of regression given a sampling of data, and then use that automatically - removing the step of a user having to manually choose what type of regression to use. This would make the predictive analysis an extremely powerful and useful tool probably, and as long as my math skills are up to the task it shouldn't be too hard to code. It would be built into the PHP half of those graphs, not the ActionScript that the graphs are physically drawn with, however, which is somewhat limited in its mathematical functions compared to some other more standard desktop languages. Regardless, it shouldn't be too much of a problem to work it out... just the time necessary to figure out the logic problems behind it.

Cool; cmafai, it's not a big deal though, so you don't really have to work on it since I'm sure it'll involve a good bit of time

But yeah, I forgot how to do it I use to have an idea, but time wasn't kind to my math skills...

Did some quick research, and what it comes down to is that nonlinear regression is a bitch. It can become extremely heavy computationally, slow, and bulky, not to mention inaccurate if outliers are factored in incorrectly. Several algorithms have been developed that will crank out some pretty nice results, but what it comes down to is that it is an inherently impossible task. As this guy named Dr. Harvey Motulsky (an expert on nonlinear curve fitting) says:

Basically, we have to church through thousands of models and compare the results of each when applied to the dataset, and then choose the best one programatically. Some models can be pruned or ignored off the bat because of the nature of their patterns, but the vast majority will have to be computed.


In other words, I can't wait to break this bitch of a puzzle. It will be done if it's the last thing I do.

Edit: Progress already! The same dude has a nice web page set up devoted to the difference between curve-fitting and regression, and the pros/cons of each. Using curve-fitting it is not necessary to choose a model, so basically you skip the step of computing those 1000s of possible set ups. Although it is less accurate, a cubic spline or lowess curve should serve our purpose well enough. That said, it's still going to be a bitch to code the algorithm that computes the curves, as they tend to be nearly impossible to convert into a usable equation form, and the math is still way over my head I believe... I'm pretty excited now though

Man, you're dedicated Haha glad you're up to the challenge, but it's not really that important

I wonder how the calculator can do it so quickly though.

Calculator only does quadratic, cubic, and logistic... and you have to choose. This system would choose the best one for you and then do it.... even if its to a power larger than say, cubic.

And I'm not dedicated - I'm messed up and find this kind of stuff fun.