Timeline for Does Mathematica have advanced indexing?
Current License: CC BY-SA 4.0
19 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 25, 2019 at 18:35 | history | edited | Szabolcs | CC BY-SA 4.0 |
deleted 12 characters in body
|
| Feb 17, 2017 at 22:00 | comment | added | Wjx | @Szabolcs Yeah, I see. Truely wierd phenomenon. This will lead to a performance drop I suppose? | |
| Feb 17, 2017 at 9:26 | comment | added | Wjx |
Yeah, Just a reminder that the selection of Select and UnitStep based solution still depends on one's application~ even in this case where Select "saved" 10000 times of calculation, it's only less than 10 times faster, so it's really inefficient in most cases.
|
|
| Feb 17, 2017 at 9:21 | comment | added | Szabolcs |
@Wjx Vectorization is always fast, but the code is hard to read and write (in Mathematica, not e.g. in MATLAB). With this code, and later with BoolEval, I tried to make it easy to read/write as well. Sadly, some bugs prevent me from doing this. Try e.g. arr = RandomReal[1, 100]; On["Packing"], then evaluate arr > 0.5. It unpacks. Why does it do that? I don't know. It doesn't even evaluate! Maybe it's a bug. Last time I complained to WRI, the support person did not understand my point and kept saying that arr < 0.5 is not correct syntax. I know, but that's no excuse for unpacking.
|
|
| Feb 17, 2017 at 9:14 | comment | added | Szabolcs |
@Wjx There are serious limitations that make this hard to implement well, e.g. Greater taking a long time to "evaluate" when it doesn't in fact evaluate at all. Things change with M versions too. Please check if it's better with the BoolEval package. I don't have time to look into this now.
|
|
| Feb 17, 2017 at 9:14 | comment | added | Wjx |
I suppose it's quite hard to solve this problem using UnitStep based solution, it's a cost of vectorization I suppose?
|
|
| Feb 17, 2017 at 9:10 | comment | added | Wjx |
Test code: cond1 = Evaluate[And @@ ConstantArray[# > 1, 1*^4]] &; dat = RandomReal[1, 1*^4]; Select[dat, cond1] // AbsoluteTiming returns {11.7397, {}} while cond = And @@ ConstantArray[dat > 1, 1*^5]; dat = RandomReal[1, 1*^4]; pick[dat, cond] // AbsoluteTiming returns {79.1119, {}}
|
|
| Feb 17, 2017 at 9:10 | comment | added | Wjx |
There are some condition where Select works better than UnitStep based solution. Notice that Select based solution can utilize the fact that True || a series of stuffs will always be True and False && a series of stuffs will always be False. so there's no need for evaluation of "a series of stuffs". While UnitStep based solution cannot easily use this fact. so when tests are significantly more than elements in list, performance can reverse.
|
|
| Jul 11, 2016 at 8:51 | comment | added | Mr.Wizard | By the way this may be relevant to you: (1922) | |
| Jul 11, 2016 at 8:47 | comment | added | Szabolcs | @Mr.Wizard Good point, thank you! | |
| Jul 11, 2016 at 8:17 | comment | added | Mr.Wizard |
Have you considered using Subtract for performance in the package code?
|
|
| Sep 29, 2015 at 8:03 | history | edited | Szabolcs | CC BY-SA 3.0 |
added 27 characters in body
|
| May 30, 2014 at 15:29 | history | edited | Szabolcs | CC BY-SA 3.0 |
added 141 characters in body
|
| Feb 11, 2014 at 20:14 | history | edited | Szabolcs | CC BY-SA 3.0 |
added 220 characters in body
|
| Feb 11, 2014 at 19:50 | history | edited | Szabolcs | CC BY-SA 3.0 |
added 142 characters in body
|
| Feb 11, 2014 at 19:37 | history | edited | Szabolcs | CC BY-SA 3.0 |
added 1211 characters in body
|
| Mar 12, 2012 at 0:33 | vote | accept | Brian B | ||
| Mar 12, 2012 at 18:07 | |||||
| Mar 11, 2012 at 21:00 | history | edited | Szabolcs | CC BY-SA 3.0 |
added 1288 characters in body
|
| Mar 11, 2012 at 20:26 | history | answered | Szabolcs | CC BY-SA 3.0 |