Talk:Tenney–Euclidean tuning: Difference between revisions
Cmloegcmluin (talk | contribs) |
Cmloegcmluin (talk | contribs) |
||
| Line 44: | Line 44: | ||
: Ah, I think I see. "Damage" may be a bit of an outdated term. It's what Paul Erlich uses in his Middle Path paper. But it means error weighted (divided) by the Tenney height, which is equivalent to the L1 norm, and so "Tenney-weighted (L1) error" is the same thing as damage. And "TE-weighted (L2) error" means error weighted by the TE height, which is equivalent to the L2 norm, so it's similar to damage. --[[User:Cmloegcmluin|Cmloegcmluin]] ([[User talk:Cmloegcmluin|talk]]) 19:04, 28 July 2021 (UTC) | : Ah, I think I see. "Damage" may be a bit of an outdated term. It's what Paul Erlich uses in his Middle Path paper. But it means error weighted (divided) by the Tenney height, which is equivalent to the L1 norm, and so "Tenney-weighted (L1) error" is the same thing as damage. And "TE-weighted (L2) error" means error weighted by the TE height, which is equivalent to the L2 norm, so it's similar to damage. --[[User:Cmloegcmluin|Cmloegcmluin]] ([[User talk:Cmloegcmluin|talk]]) 19:04, 28 July 2021 (UTC) | ||
:: Corrections: | |||
:: * The term damage is not outdated. | |||
:: * My quotations from the article are out of date. They show "L1" and "L2" in parenthesis, which implies that they "Tenney-weighted error" is the same thing as "L1 error" and that "TE-weighted error" is the same thing as "L2 error". Those statements would both be incorrect. "Tenney-weighted error" is the "T1 error" and "TE-weighted error" is the "T2 error". I see that as of Jan '22, Flora has fixed this by removing the parenthesis, so that it's clear that the L1 error is being Tenney-weighted (to become T1) and the L2 error is being Tenney-weighted (to become T2). My previous comment did not reflect this understanding, stating that Tenney height was the L1 norm (it's actually the T1 norm) and that TE height was the L2 norm (it's actually the T2 norm). Erlich's "damage" is the T1-weighted absolute value of error though, so it is closely related to T2-weighted absolute value of error. | |||
:: * But "error" should not be replaced with "damage" as I'd suggested. Damage is a weighted abs val of error. So the article states things with respect to "error" correctly. | |||
:: --[[User:Cmloegcmluin|Cmloegcmluin]] ([[User talk:Cmloegcmluin|talk]]) 23:53, 5 March 2022 (UTC) | |||
== "Frobenius" tuning == | == "Frobenius" tuning == | ||