Template talk:Floating-point

Remove G.711 from floating-point formats

As I understand it, G.711 is a specific encoding that is only relevant for telephony audio signals. I don't think it belongs with the list of other floating point formats, which are all general purpose and not domain specific. I don't think domain-specific encodings, even if they happen to use a floating-point-like format, should be included in the template. — Preceding unsigned comment added by Azmisov (talk • contribs) 00:15, 6 January 2022 (UTC)

This wasn't done and it shouldn't be, for two basic reasons : - It doesn't harm the content at all - If removing domain-specific encodings would be appropriate, then you should also remove proprietary formats. This would make the template rather useless and seriously biased towards one family of encodings.

Consideration on Posit numbers

While it looks like there is some discussion about Posit being floating point representations, I suggest that the arguments against including Posits in the list be discussed here. There could be some confusion with other uses of the Posit name, but the works of John Gustafson clearly define Posit arithmetic as an alternative to IEEE754. See for example http://www.johngustafson.net/pdfs/BeatingFloatingPoint.pdf that states explicitly "A new data type called a posit is designed as a direct drop-in replacement for IEEE Standard 754 floating-point numbers (floats)." The main differences are that Posits are "tapered" (with less relative precisions for high powers of two) and 2's exponent is encoded in 2 different fields (regime and exponent). — Preceding unsigned comment added by Tochinet (talk • contribs) 11:32, 29 October 2024 (UTC)

John Gustafson said many things that are incorrect. Posits cannot be a drop-in replacement for floating-point numbers. This will break algorithms (like 2Sum) that assume a floating-point format. Please do not cite anything from John Gustafson. He just wants to sell his Posit format. — Vincent Lefèvre (talk) 13:31, 29 October 2024 (UTC)

While I understand that you may consider Posits an "inferior" format to IEEE754, and I appreciate the discussion, I can guarantee you that Posit _can_ be a replacement for floats. I find very unfortunate that the discussion becomes opinionated. I have no connection with Gustafson and the [Posit Standard] may not meet your expectations of what a "true" floating point format is, but nevertheless, 2Sum is a bit light argument for rejecting it altogether. For small microcontrollers (ESP8266 or even AVR 8-bit), there is a real interest to have a floating-point 8-bit or 16-bit format. Several papers such as [The Accuracy and Efficiency of Posit Arithmetic] show that tey definitely can and are compared to IEEE754.

In addition to my comment above, I believe you should look again at the [[1]] page. I really don't want to offend, I'm only an occasional contributor, and I won't engage in a revert war, but however bad you think they are, Posits are really innovative, they got praised by many people, got featured on [IEEE Spectrum] etc. Could I suggest you rather would post some of your critiques on the Unum discussion page, where Posits are described ? I would see no issue if the Posit reference was moved to the "Alternatives" instead of "others" in the template, but Posits don't deserve being silenced. On the contrary, I believe that adding as well [Takum arithmetic] would be a useful addition, because it is innovative. Tochinet (talk) 12:37, 30 October 2024 (UTC)