Anonymous user
Talk:Kahan summation: Difference between revisions
Another note to the task
(Another note to the task) |
|||
Line 94:
Thanks Sonia. I am still digesting your suggestion, but just on first reading, (and little cogitation I admit), the solution is not "self scaling" in any way, for different precisions. The current task floating point examples might self scale in say Python, but do not in J because of their "fuzzy comparison" (a problem with the task description and not a slur of J by the way). --[[User:Paddy3118|Paddy3118]] ([[User talk:Paddy3118|talk]]) 05:18, 14 December 2014 (UTC)
==Task 2==
I suggest to rewrite this task from scratch, giving in the task description 7 or 10 64 bit floating point numbers (or more if generated with a portable random generator), and show the results with a basic summation algorithm versus doing it with Kahan. And show in some ways that the second loses less precision. Nothing more.
| |||