The Unbearable Lightness of Unum?

[fuzzy wozzy]

Unum is revolutionary in its concept, to rekindle the effort started by interval math decades ago;
daring in its ambition to bring accuracy and exactness of floating point calculations to every desktop
computer and supercomputer, both from the software and hardware perspective.

The Unum book is praised as being important and well written, also clearly written enough that even high school
students can follow, which means that if they're learning Java or C++, they can write it in those languages easily,
at least this is the impression one gets from reading the reviews about it.

Yet, it turns out to be tough enough to stymie the best-intentioned efforts of a Shenturian?
How could this be? And how to explain the seeming lack of interest or enthusiasm from other Shenturians,
at least judging from the lack of response to the things posted about it?

Of course, some may be reading the book and writing code in Shen, language designed for easy use yet
as powerful, if not more, as any language in use today. The high capabilities of the Shenturians have been
demonstrated time and again with each release of a new porting of Shen to XYZ language.
There is no questioning that anyone who can do that could no doubt do Unum in Shen, too.

And it isn't as if floating point calculation affects only a few people, it affects EVERYBODY.
I remember years back when consumers were returning their new computers for an exchange because
the news media reportedd that some of the computers were not doing the floating point math correctly,
people did not want to have a computer with such deficiency, and not just gamers or programmers.

Now more than ever, with online financial transactions and investments, this would become more of a
relevance, not less. Then, it's only natural to find it puzzling that, with a good book and a good language,
and a gathering of brilliant minds willing to work hard to solve a worthy problem, the progress seems to be
awful slow, but some say patience is a virtue, so there's that as well...

[Neal Alexander]

I assume It's not necessarily a lack of interest, but that people don't have anything relevant to contribute to the conversation. How many hackers are an expert on the low level implementation details of floating point? My eyes glaze over when I look through https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html. Any comparison with Unum requires a person to be intimately familiar with competing number representations. 

A software implementation is going to be slow. If you don't care about speed, you already have a variety of solutions available in bignum or symbolic computation packages. It's unclear, to the ignorant, why Unum is better. However, It looks like there is going to be a conference debate this month, between Kahan and Gustafson, which should be interesting.

Finally, it may be wrong to use software based arithmetic for Shen's default number type. You are better off implementing it via native hardware instructions, since the user has no simple way of generating them from within Shen, and function call overhead is too high if you care about performance. If a user needs slow, but high precision, software numbers he can just link against a third party library. 

[Willi Riha]

Please, stop this thread!

We don't need two about unums and Mathematica.

fuzzy wozzy,  you can only comment about the book and the ease of translating the esoteric Mathematica code into any other language if you have had a go yourself.

A C -translation seems difficult, as this language lacks lists. It would be alright with bit operations. 

As far as other languages, you obviously have no clue!

On the other hand, ordinary (portable) Shen does not support bitwiise operations - so forget about it.

Professional Shen can!

Yes, you are right, floating-point computations have always been problematic. If  unums are the answer to a maiden's prayer, willl have to be seen..

[Mark Tarver]

On Sunday, 10 July 2016 04:08:46 UTC+1, Willi Riha wrote:

Please, stop this thread!

I'd tend to agree; the number of people in this group who can profitably contribute to the unum discussion is very limited and Willi is one of the few who has bought the book and read it.  So one thread is enough to be going on with.   

Mark

[fuzzy wozzy]

The free Shen is still generous enough that if you're working with a 10 million-digit binary number,
it would take less than 0.24 seconds to do leftshift on it. But I might do it in OCaml since you've
done most of it in Shen already, not wanting to duplicate the effort.

Unum 2.0 might be more interesting to do it Shen.

[Mark Tarver]

I understand Willi is writing up complex numbers; so if you want to do it - go ahead.

Mark

[Antti Ylikoski]

Dear Mark:

I wrote one version of the complex numbers package, making an abstract

datatype out of the complex numbers.  Willi wrote a number of

negative, and mostly wrong, comments on it.

I modified the complex numbers package according to Willi's comments,

and wrote a second complex numbers package.  Willi again authored a

second file, a number of negative comments, mostly wrong.

I attach Willi's second comment file into this Shen group entry with

my own notes, in order for the group to see what Willi wrote.  That

will be self explanatory.

Is it the case that, now, the complex numbers package that I wrote has

been discarded, and Willi is writing a complex package which will

instead be a part of SP?

The idea of a complex package was mine; how much of the said complex

work by Willi is based on my work?

yours, A. J. Y.

HELSINKI

Finland, the E.U.

[Mark Tarver]

I confess that, being  so busy, I haven't really looked closely at his code and your most recent code. Willi has been in charge of the maths library since its inception in 2010 because Willi has a Ph.D. in maths and 30 years teaching experience in computing.  His maths lib is, IMO, the best and most thoroughly documented submission to the OS library to date.   So when you sent me your submission, I simply passed it to him as the man in charge of that area.  

From my cursory glance, I think that the model Willi proposed was closely based on the model for representing rational numbers that I sketched out in TBoS.  You followed a different course.  Obviously from my position I'd think Willi's approach was the way to go and I explained this approach to him in person quite few moons ago.   He just got around to implementing it.

Willi was trying to be constructive but I really don't think that the content of private email between you and him should have ever been raised in the first place, and certainly not in the context of his original mail on unums because private email should remain that way.   I haven't any axes to grind here,  You are certainly entitled to confidentiality.   I also suggest to the mods that in future unless both parties agree to the disclosure of private email, that private differences should not be aired publicly.  

 Whatever the rights and wrongs of this are, I don't believe Willi plagiarised your code because, for one thing, AFAIK the data structures he used to represent complex numbers were very different from yours - that was the heart of the disagreement. Any resemblances beyond that arise because many of the mathematical definitions are quite standard.   

But in all we are talking about a small amount of not too difficult code and we need to keep a sense of proportion. But in future I'll make certain that any code submissions from you or anybody stay with me and are treated as confidential.

Mark 

--
You received this message because you are subscribed to the Google Groups "Shen" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qilang+un...@googlegroups.com.
To post to this group, send email to qil...@googlegroups.com.
Visit this group at https://groups.google.com/group/qilang.
For more options, visit https://groups.google.com/d/optout.

[Antti Ylikoski]

Willi showed to me his rational numbers implementation.

It is based on absvectors, which are not a part of the Shen type

system.

Therefore, Willi's complex numbers package most probably is not type

secure.

The complex numbers package that I authored, handles complex numbers

as pairs, and therefore it is type secure.

yours, A. J. Y.

HELSINKI

Finland, the E.U.

[Mark Tarver]

You really have to take this up with Willi if you want to take it further.   This is really an overspill of a private correspondence involving a difference between two people.  

Mark

[Antti Ylikoski]

Thank you for some wise comments, Mark.

I shall not commence or continue any argument, with anybody, in any

context, and I feel that I owe an apology to many individuals, if it

has begun to look like that, there is a fencing fight in the Shen

group.  That is not in the spirit of science.

But my statement about the complex library with absvectors, or with

pairs, is a scientific statement, it is not an attempt to argue with

anyone in any manner.

I mean, the type system is an important part of the Shen system; I

have previously written that the Shen types are an excellent help for

the programmer who is attempting to make something truly complex.

I even have been planning to write a very positive article about Shen

in the ICT press.  The Shen system deserves such treatment.

The absvectors are not a part of the Shen system.  But it is easy to

create a type secure complex package with pairs.  That is all I want

to say, this has not been written to argue with anybody.  It is a

purely scientific statement.

yours, A. J. Y.