Talk:Verify distribution uniformity/Naive: Difference between revisions

Content added Content deleted

Inline

Revision as of 21:26, 24 September 2009

We really ought to use a chi-squared test for this, as that can be made self-calibrating. After all, we've got the tools for calculating the Gamma function, needed for generating the related distribution for a single random variable. Too early in the morning for heavy math for me though… —Donal Fellows 06:16, 8 August 2009 (UTC)

After reading your link, and being up early creating the task in the first place (I'm in Bristol), I also would not want to tackle the maths ;-)

Please, feel free to add another task to run a chi-square test on the results of Seven-dice from Five-dice, but write the task in such a way that enough languages would be able to compute it if possible. (But then, if mathematica or R have a built-in function, shouldn't they be able to shine)? --Paddy3118 07:27, 8 August 2009 (UTC)

And why shouldn't they shine at something they're good at? —Donal Fellows 11:56, 8 August 2009 (UTC)

Now over at Verify Distribution Uniformity with Chi-Squared Test‎. Go knock yourselves out. ;-) —Donal Fellows 12:15, 9 August 2009 (UTC)

What is Delta?

It would be nice if the interpretation of the delta parameter were more clearly specified. I don't feel comfortable improvising. —Dennis Furey 21:54, 8 August 2009 (UTC)

"...check bin counts are within +/- delta % of repeats/bincount" (From the Python example).

I kinda knew that people with more experience probably wouldn't do it that way, (See the Chi-square comment above); but thought that if you took a fixed sample of a million, any fitness metric should be able to be translated into this form, so went with it. I have no idea of what is good-enough, and also didn't want to parrot some figure of fitness that I did not understand. --Paddy3118 05:54, 9 August 2009 (UTC)

Of course a random number checker like this will report a perfect random distribution if your "RNG" returns the sequence "1234567123456712345671234567...". BTW, shouldn't the function also be given info about what values are to expected? Because if a random number generator intended to emulate a normal six-sided dice actually returns a flat distribution of numbers from 1 to 7 or from 1 to 5, it's certainly not very good :-) --Ce 21:26, 24 September 2009 (UTC)

Renamed

I renamed this task so that I can put in a task (or tasks) that does a more sophisticated job and which will give various languages' statistics support a better workout. —Donal Fellows 11:47, 9 August 2009 (UTC)

@@ Line 10: / Line 10: @@
 :"...check bin counts are within +/- delta % of repeats/bincount" (From the Python example).
 :I kinda knew that people with more experience probably wouldn't do it that way, (See the Chi-square comment above); but thought that if you took a fixed sample of a million, any fitness metric should be able to be translated into this form, so went with it. I have no idea of what is good-enough, and also didn't want to parrot some figure of fitness that I did not understand. --[[User:Paddy3118|Paddy3118]] 05:54, 9 August 2009 (UTC)
+:: Of course a random number checker like this will report a perfect random distribution if your "RNG" returns the sequence "1234567123456712345671234567...". BTW, shouldn't the function also be given info about what values are to expected? Because if a random number generator intended to emulate a normal six-sided dice actually returns a flat distribution of numbers from 1 to 7 or from 1 to 5, it's certainly not very good :-) --[[User:Ce|Ce]] 21:26, 24 September 2009 (UTC)
 == Renamed ==