Re: RangeExtra 10^-20
在 Sat, 25 Apr 2009 23:40:03 +0800,dsimcha dsim...@yahoo.com 写道: == Quote from dsimcha (dsim...@yahoo.com)'s article == Quote from Simen Kjaeraas (simen.kja...@gmail.com)'s article dsimcha wrote: RangeExtra version 10^-20 is officially out. It consists of a small and hopefully growing number of ranges that didn't make it into the new Phobos that I've gotten working reasonably well and I feel eventually belong in Phobos. Docs / What's there: http://cis.jhu.edu/~dsimcha/rangeextra.html Code: http://dsource.org/projects/scrapple/browser/trunk/rangeextra/rangeextra.d License: Dual licensed, Phobos license or BSD (Tango style). I had hoped reindex would give me access to true 'random' access ranges, but it seems this does not work: auto b = reindex!( uniform( 0, this.length ) )( [0,1,2,3,4].dup ); :p On the other hand, I really like the idea of a with-replacement random sampling primitive. This sounds like it belongs in RandomCover. Maybe Andrei can extend RandomCover (which already samples w/o replacement) to include a replacement parameter. RandomCover w/ replacement == false would do what it does now. If replacement == true, it would be an infinite range that samples with replacement. Here's a first attempt: I took Andrei's RandomCover, fixed bug 2865 (This version of RandomCover passes basic statistical tests for uniformity of the distribution over the permutation space), and added a template overload to allow sampling with replacement. It's under the Phobos license to make it as easy as possible to add back to Phobos. I'll file an enhancement request. http://dsource.org/projects/scrapple/browser/trunk/rangeextra/randomcover.d Looking forward to it! -- 使用 Opera 革命性的电子邮件客户程序: http://www.opera.com/mail/
Re: RangeExtra 10^-20
dsimcha wrote: RangeExtra version 10^-20 is officially out. It consists of a small and hopefully growing number of ranges that didn't make it into the new Phobos that I've gotten working reasonably well and I feel eventually belong in Phobos. Docs / What's there: http://cis.jhu.edu/~dsimcha/rangeextra.html Code: http://dsource.org/projects/scrapple/browser/trunk/rangeextra/rangeextra.d License: Dual licensed, Phobos license or BSD (Tango style). I had hoped reindex would give me access to true 'random' access ranges, but it seems this does not work: auto b = reindex!( uniform( 0, this.length ) )( [0,1,2,3,4].dup ); :p
Re: RangeExtra 10^-20
Simen Kjaeraas wrote: dsimcha wrote: RangeExtra version 10^-20 is officially out. It consists of a small and hopefully growing number of ranges that didn't make it into the new Phobos that I've gotten working reasonably well and I feel eventually belong in Phobos. Docs / What's there: http://cis.jhu.edu/~dsimcha/rangeextra.html Code: http://dsource.org/projects/scrapple/browser/trunk/rangeextra/rangeextra.d License: Dual licensed, Phobos license or BSD (Tango style). I had hoped reindex would give me access to true 'random' access ranges, but it seems this does not work: auto b = reindex!( uniform( 0, this.length ) )( [0,1,2,3,4].dup ); :p If I had to take a guess, I'd say it was because uniform isn't defined in the context of reindex. Maybe this: import std.random; void foo() { auto b = reindex!((i){ return uniform(0,5); })( [0,1,2,3,40.dup ); } I don't think there's any way to get the length. -- Daniel
Re: RangeExtra 10^-20
== Quote from Simen Kjaeraas (simen.kja...@gmail.com)'s article dsimcha wrote: RangeExtra version 10^-20 is officially out. It consists of a small and hopefully growing number of ranges that didn't make it into the new Phobos that I've gotten working reasonably well and I feel eventually belong in Phobos. Docs / What's there: http://cis.jhu.edu/~dsimcha/rangeextra.html Code: http://dsource.org/projects/scrapple/browser/trunk/rangeextra/rangeextra.d License: Dual licensed, Phobos license or BSD (Tango style). I had hoped reindex would give me access to true 'random' access ranges, but it seems this does not work: auto b = reindex!( uniform( 0, this.length ) )( [0,1,2,3,4].dup ); :p Very cool idea. Nothing like what I had in mind as a use case for Reindex, but half the fun of generic programming is seeing stuff get used for things you hadn't anticipated. I spent some time trying, and for some reason I can't for the life of me get this example to work, or even the one where the length is hard-coded. It may be some obscure compiler bug, because I'm just getting weird inscrutable error messages and I can't really figure out where they're coming from. On the other hand, I really like the idea of a with-replacement random sampling primitive. This sounds like it belongs in RandomCover. Maybe Andrei can extend RandomCover (which already samples w/o replacement) to include a replacement parameter. RandomCover w/ replacement == false would do what it does now. If replacement == true, it would be an infinite range that samples with replacement.
Re: RangeExtra 10^-20
== Quote from dsimcha (dsim...@yahoo.com)'s article == Quote from Simen Kjaeraas (simen.kja...@gmail.com)'s article dsimcha wrote: RangeExtra version 10^-20 is officially out. It consists of a small and hopefully growing number of ranges that didn't make it into the new Phobos that I've gotten working reasonably well and I feel eventually belong in Phobos. Docs / What's there: http://cis.jhu.edu/~dsimcha/rangeextra.html Code: http://dsource.org/projects/scrapple/browser/trunk/rangeextra/rangeextra.d License: Dual licensed, Phobos license or BSD (Tango style). I had hoped reindex would give me access to true 'random' access ranges, but it seems this does not work: auto b = reindex!( uniform( 0, this.length ) )( [0,1,2,3,4].dup ); :p On the other hand, I really like the idea of a with-replacement random sampling primitive. This sounds like it belongs in RandomCover. Maybe Andrei can extend RandomCover (which already samples w/o replacement) to include a replacement parameter. RandomCover w/ replacement == false would do what it does now. If replacement == true, it would be an infinite range that samples with replacement. Here's a first attempt: I took Andrei's RandomCover, fixed bug 2865 (This version of RandomCover passes basic statistical tests for uniformity of the distribution over the permutation space), and added a template overload to allow sampling with replacement. It's under the Phobos license to make it as easy as possible to add back to Phobos. I'll file an enhancement request. http://dsource.org/projects/scrapple/browser/trunk/rangeextra/randomcover.d
RangeExtra 10^-20
RangeExtra version 10^-20 is officially out. It consists of a small and hopefully growing number of ranges that didn't make it into the new Phobos that I've gotten working reasonably well and I feel eventually belong in Phobos. Docs / What's there: http://cis.jhu.edu/~dsimcha/rangeextra.html Code: http://dsource.org/projects/scrapple/browser/trunk/rangeextra/rangeextra.d License: Dual licensed, Phobos license or BSD (Tango style).