Friday, September 10, 2010

New Algorithm (again)

Howdy all!

Once again, I've changed the algorithm of the new feature (soon to be released) that will predict what you need to order for inventory. After further testing, I determined that it could be influenced by a mass movement way too much.

I now focus mainly on the Average movement. However, I do still take Standard Deviation into account. If there is a huge standard deviation (most likely a mass movement of literature), LRO will scale down the calculation of the average. So it will not be thrown way off. For those interested, here is the actual formula simplified:

Threshold = Avg * Calc

Calc =
If StDev > Avg Then
-----If Avg/StDev < .7 Then
----------Calc = Avg/StDev
----------Calc = 1
-----End If
-----Calc = 1
End If

StDev will be greater than the Avg on mass movements. Avg/StDev will determine just how large of a deviation it was. The closer to .5 it is, the larger the deviation from the average. If it is below .7, then the Avg is multiplied by it, thus reducing the Avg accoring to how large of a deviation it was.

Average is muliplied by the sensitivity (Low=1, Normal=2, High=3) so it can be used to look ahead up to 3 months. Sensitivity also changes the .7 value (Low) to .65 (normal) and .6 (High). This means that higher sensitivity will allow the deviation to affect the average more (it will be less likely to reduce the avg due to Standard Deviation). From my tests, this algorithm is even more accurate.

I hope I didn't hurt anyone's head with all this! ;-)

PS - Testing is going well!

No comments:

Post a Comment