Timeline for Can Mathematica's random number generation be improved?
Current License: CC BY-SA 3.0
34 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 23, 2018 at 21:37 | answer | added | Henrik Schumacher | timeline score: 5 | |
| Apr 13, 2017 at 12:56 | history | edited | CommunityBot |
replaced http://mathematica.stackexchange.com/ with https://mathematica.stackexchange.com/
|
|
| Oct 6, 2016 at 7:33 | history | edited | dr.blochwave | CC BY-SA 3.0 |
deleted 455 characters in body
|
| Feb 24, 2015 at 3:36 | history | edited | JEP | CC BY-SA 3.0 |
deleted 2 characters in body
|
| Feb 24, 2015 at 1:34 | history | edited | JEP | CC BY-SA 3.0 |
Added an explanation for the motivation of this question and a short discussion of the 3 answers.
|
| Feb 20, 2015 at 23:45 | comment | added | ciao | @jep Then you use the same technique where the parameter RVs are defined by your inputs. Pretty basic stuff. | |
| Feb 20, 2015 at 23:07 | comment | added | dr.blochwave | Ok, I understand now, thanks for clarifiying | |
| Feb 20, 2015 at 22:56 | comment | added | JEP | My input parameters come from a simulation. They don't have a known parametric distribution. | |
| Feb 20, 2015 at 22:31 | comment | added | dr.blochwave | @JEP why not? The example provided by rasher below gives the same histogram as MMA or C++, and is very quick to generate and draw from (especially for very large draws of $>10^{7}$) | |
| Feb 20, 2015 at 22:23 | comment | added | JEP | If you are assuming a distribution for the input parameters on the parametric generation then you are not generating the random numbers that I need. | |
| Feb 20, 2015 at 22:12 | comment | added | ciao | @blochwave: Added... | |
| Feb 20, 2015 at 22:01 | comment | added | dr.blochwave | @rasher look forward to seeing it, especially if it beats the C++! | |
| Feb 20, 2015 at 22:00 | comment | added | ciao | @blochwave: That 200ms was on a netbook, 6x faster than the R example on same - but it's basic MMA thinking: Use ParametricMixture to get dist, PDF (or CDF depending on dist type), use PDF to drive RandomChoice (or CDF for inversion). Will add binomial example to my answer when time permits. | |
| Feb 20, 2015 at 21:56 | history | edited | dr.blochwave | CC BY-SA 3.0 |
improved the layout of the question, easier to read, formatted the C code block as C not MMA.
|
| Feb 20, 2015 at 21:47 | comment | added | dr.blochwave | @rasher out of interest, what was your code for the parametric generation of 10^6 binomial variates? 200ms is close to the performance of the C++ code (150ms) | |
| Feb 20, 2015 at 19:55 | review | Close votes | |||
| Feb 20, 2015 at 23:22 | |||||
| Feb 20, 2015 at 19:28 | comment | added | ciao | @AlbertRetey well said. | |
| Feb 20, 2015 at 19:27 | comment | added | ciao | Seeing as intelligent use of Mathematica provided faster generation than the C and R examples below, I don't think it's WRI that needs to do any fixing. Perhaps the Experimental`AngerManagement package needs a spin... | |
| Feb 20, 2015 at 17:10 | comment | added | JEP | Sorry about that. I wasn't trying to trick anyone into reading anything. The first line in my post is: "Performance on Random Number generation is intolerable." I'm not sure why you kept reading if you don't care about performance. WRI can fix it or not. | |
| Feb 20, 2015 at 15:50 | comment | added | Albert Retey | If you feel hostility I think that might have to do with the fact that you chose to use a quite aggressive title, which also isn't in line with your actual question. You might realize that many readers would be seriously concerned about broken random number generators while they are neither surprised nor worried to hear that Mathematica is slow for this kind of task. They might (as I did) feel being tricked into reading your (lengthy) question in the first place... | |
| Feb 20, 2015 at 6:25 | vote | accept | JEP | ||
| Feb 19, 2015 at 5:09 | history | tweeted | twitter.com/#!/StackMma/status/568276258372620289 | ||
| Feb 18, 2015 at 22:32 | answer | added | JEP | timeline score: 7 | |
| Feb 18, 2015 at 15:23 | history | edited | JEP | CC BY-SA 3.0 |
deleted 32 characters in body
|
| Feb 18, 2015 at 15:15 | history | edited | JEP | CC BY-SA 3.0 |
improved formatting
|
| Feb 18, 2015 at 15:13 | comment | added | wolfies | Yes - but your question is messy. A good question simplifies the problem down to 1 or 2 lines (if possible), and it is certainly possible to do so here. By contrast, your question appears lazy, because it just dumps lines and lines of code from whatever you were doing, without honing it down to the heart of the matter. 2 lines: no more. Further, your title asks if it can be FIXED, which infers that it is broken. But your question is not about quality, nor do you compare the quality of pseudorandom drawings generated, nor does your post suggest anything is broken. | |
| Feb 18, 2015 at 13:58 | history | edited | JEP | CC BY-SA 3.0 |
deleted 35 characters in body
|
| Feb 18, 2015 at 13:04 | comment | added | JEP | f should not have been there. Other than that ints, reals, and mus were all used as arguments to the distributions. I am detecting hostility here which I do not understand. I don't understand the point of comments that don't address the question. My point is that if you can generate random variates all at once the performance is comparable to C so it is false that "of course compiled C is faster than Mathematica." There are many simulation problems in which random numbers with different parameters are needed as in the examples I provided. | |
| Feb 18, 2015 at 12:53 | comment | added | JEP | I was not aware that symbolic calculations have anything to do with random number generation? ?? | |
| Feb 18, 2015 at 10:57 | comment | added | wolfies |
I don't get the point either. If you want to do a comparison, get it down to 2 lines of code. Do you really need to define num, ints, reals, mus, f, mf, rvs, gf ... just to do a timing test? Talking of efficient, who has time to look through all that?
|
|
| Feb 18, 2015 at 8:33 | answer | added | dr.blochwave | timeline score: 40 | |
| Feb 18, 2015 at 7:43 | answer | added | ciao | timeline score: 37 | |
| Feb 18, 2015 at 6:58 | comment | added | ciao | What's your point? Of course compiled C is faster than Mathematica. So call your C from it if that kind of speed is critical for this simplistic stuff. On the other hand, I'll wait patiently while you use your C code to evaluate complex probabilistic equations symbolically... | |
| Feb 18, 2015 at 6:25 | history | asked | JEP | CC BY-SA 3.0 |