Hacker News new | past | comments | ask | show | jobs | submit login

I sort of like the idea of avoiding exactly representable floats (you typed 4 but maybe you get 4-ε) to remind programmers that errors creep into most expressions and compound, and tight error bounds require numerical error analysis.



I wish we hd float hardware that tracked epsilon for you. Floats would be two values with epsilon being updated on each operation.

Then a comparison would take epsilon into account instead of just a bit match.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: