[WP34s] A set of new quantile functions

[Dieter]

(05-26-2014 09:36 AM)fhub Wrote:  That's a bit strange in my opinion:
First you complain about the illogical order of arguments for IBeta a few weeks ago, then Pauli has changed that (and it was good so), and now you provide a new set of code for distribution functions with exactly the same old (illogical) order of arguments?

Yes. The essential point was not the order of the Iβ arguments, but the fact that the function worked different from what the manual said. This has been corrected. Iβ now works "as advertized", which indeed is more logical.

But there is another point. The beauty of the German language has gifted us with the term "Die normative Kraft des Faktischen". What do you think, how many 34s are currently using the latest firmware? Five percent? Ten? That's why I assumed the previous Iβ implementation. All others may add a simple stack shuffle command.

Quote:It would really be nice to have a (quite) bugfree WP34s version finally

That's why I humbly proposed these three routines. The previous versions sometime threw errors or required inacceptable execution times.

What about you? Your TVM solver has made it into the 34s and added some essential functionality. Great. But what about another project? While I worked on the quantile functions I discovered both the Iβ bug as well as the L.R. problem. Both have been corrected – and the 34s became a bit better. This is a community effort – "ask not what the 34s can do for you – ask what you can do for the 34s project". :-)

Quote:(and I don't care whether any high-level function is accurate to 37 or 'only' 36 digits internally ).

It's more like 30...34. There are limitations in the algorithm and the functions it uses as well as in the hardware. But I think 30 or 32 digits are good enough. If you want all 34 in DP mode, having these functions coded in C is the only way. ;-)

Dieter