[RESOLVED] float precision question
I'm writing some software to generate nonstandard odds for Keno in an online casino setting.
To bring out whole number results, I'm using a non-gaussian smoothing function I wrote, that compresses and massages a hyperbolic set of numbers into something useful. My smoothing function runs multiple times on the data and maintains precision up to 99.9999999991% or so after the 3rd iteration.
In Keno, certain possibilities are extremely remote (in the quintillions) and naturally generate payouts in the billion-dollar range. In order make the game more playable, wildly high paybacks on the outsides of the hyperbola are reduced and refactored toward center based on an initial limiter and compression factor, their value is divided by the difference in odds between the more and less probable outcomes. That division and refactoring is extremely sensitive: Slight modifications in the 16th decimal place can mean hundreds of dollars difference when the smoothing is performed several times.
The Flash function works like a charm; better than a charm, actually. The problem is, this function MUST sync with the exact same function in PHP, where the REAL payouts are being calculated. On my system, floating point precision in PHP is 3 places less than it is in Flash. Although precision is supposed to be 53-bit for Number and float types in both languages, there is a mismatch directly caused by the three missing decimal places in PHP... the result being that Flash is up to 1% more precise for each run of the function.
I'd like to keep the Flash level of precision in PHP -- I'm experimenting with the BigNumbers classes on that side -- what I'm wondering is, does anyone have a way to mod the Number type in Flash, should it become necessary, to arbitrarily lower the floating point precision without a huge amount of multiplying, rounding and dividing?
A Flash Developer Resource Site
Reply With Quote
