Handful of algorithm and convention fixes

webmachinelearning / webnn

🧠 Web Neural Network API

https://www.w3.org/TR/webnn/

Other

397 stars 48 forks source link

Handful of algorithm and convention fixes #672

Closed inexorabletash closed 7 months ago

inexorabletash commented 7 months ago

Bug fix: Shapes should not be broadcast for logical "not" op.
Bug fix: Validate transpose()'s permutation as a list, not operand.
Conventions: Use "be" and "let" rather than "denote" and "store".
Conventions: Use "otherwise" instead of "else" in steps.
Conventions: Use "for each" not just "for" when iterating.

Preview | Diff

inexorabletash commented 7 months ago

Some minor fixes noted along the way. All editorial - either purely wording changes, or where the algorithm was clearly wrong but the intent was obvious. Please take a look @huningxin and @fdwr - thanks!

inexorabletash commented 7 months ago

And re: "else" vs. "otherwise" - Infra talks about this, but we also used "otherwise" at least 2x as frequently as "else".

fdwr commented 7 months ago

And re: "else" vs. "otherwise" - Infra talks about this, but we also used "otherwise" at least 2x as frequently as "else".

Ah yes, I remember those BASIC keywords well, IF, THEN, and OTHERWISE 😅... Alrighty, if our spec already uses otherwise 2x more than else, then consistency argument win.