List Commands Library for 50g

[DavidM]

(03-17-2018 11:19 PM)pier4r Wrote:  In the worst case if you have a spare 50g, you can copy/adapt my code and get the results faster.

I'm attempting to run your code on an emulated 50g. The first run was with all the defaults as you set them in your earlier program. I used the 1.1.3b4 library, and it still had similar results (in one run, standard RAND method was 583, 4096 for LSHUF).

I then changed the "list of elements" (in both places) by adding a value of 50 to make it now have 5 elements, and changed the number before RAND in the "produce a list of N picks of elements that is the goal list" to 5 so that it would include all 5 elements in its picks. I left the length of the goal at 6. This time, both methods went "to the wall" at 15625 attempts. So neither method seemed to produce the targeted list. Did I neglect to change something appropriately? Granted, this was only 1 loop iteration so it may not be as meaningful.

(03-17-2018 11:19 PM)pier4r Wrote:  One more question. Why did you use the rightmost digits of the random number, if you are going to draw another random number afterwards? I mean, any particular reason?

The choice I made wasn't specifically related to using the rightmost digits, but rather using a method that employed integer operations as opposed to floating point math. I knew that using MOD against a seed would allow me to quickly come up with a result in the range that I needed. It just happens that using that particular method inherently derives its result from the least significant digits in the seed. That method is now suspect due to the apparent "random qualities" of the bits in that area of the seed.

So the question remains: does shifting the seed give enough improvement to continue using this method, or should some other method be deployed?

Ironically, the main reason I opted to use the internal RAND functionality was because I thought it would keep me from having to come up with an alternate version that I would then have to validate. It just turns out that the method I chose to convert the resulting seed to the final number I need is now shown to be questionable.