I still would not exclude that your implementation of memoization can be improved somewhat (will check someday).

My little program reports speed in millions of integers per second. So higher values are better.

Do you still think our results are inconsistent?

to32: 16.384 32: 10.923 32Fast: 13.107 32Memo: 9.362 to64: 10.923 64: 10.923 64Fast: 8.192 64Memo: 9.362
1 to32: 16.384 32: 4.369 32Fast: 10.923 32Memo: 5.958 to64: 9.362 64: 0.809 64Fast: 8.192 64Memo: 5.958
2 to32: 16.384 32: 2.849 32Fast: 9.362 32Memo: 4.096 to64: 8.192 64: 2.114 64Fast: 7.282 64Memo: 3.855
3 to32: 5.461 32: 1.82 32Fast: 8.192 32Memo: 2.849 to64: 7.282 64: 1.337 64Fast: 5.958 64Memo: 2.731
4 to32: 13.107 32: 1.092 32Fast: 6.554 32Memo: 2.185 to64: 6.554 64: 0.728 64Fast: 5.041 64Memo: 1.928
5 to32: 13.107 32: 0.607 32Fast: 4.681 32Memo: 1.598 to64: 5.461 64: 0.428 64Fast: 4.369 64Memo: 1.394
6 to32: 7.282 32: 0.324 32Fast: 3.449 32Memo: 1.04 to64: 4.369 64: 0.224 64Fast: 3.121 64Memo: 0.923
7 to32: 3.277 32: 0.168 32Fast: 2.427 32Memo: 0.655 to64: 2.849 64: 0.115 64Fast: 2.427 64Memo: 0.516
8 to32: 1.725 32: 0.085 32Fast: 1.68 32Memo: 0.383 to64: 1.598 64: 0.058 64Fast: 1.82 64Memo: 0.298
9 to32: 0.851 32: 0.043 32Fast: 1.024 32Memo: 0.21 to64: 0.95 64: 0.029 64Fast: 1.285 64Memo: 0.172

>I’d be curious to see how you implement it.

In Java, it is a bit a hard. In C++, you just use an array of structures.

On my laptop, BTW, naive encodings are the fastest, not fast32, fast64 versions.

Really? Can you share the output you get?

Please use the -server flag with java, otherwise numbers tend to be all over the place.

I would have put all the data in a single contiguous array (maybe better from a perspective of caching).

I’d be curious to see how you implement it.

I also wonder if results obtained for Java also hold for C++.

I agree. I might hack something together.

I feel that the editor would have no ammo once he receives the reviews from the reviewers

The objectivity of the editor is important especially because reviewers are rarely unanimous. Editors effectively act as super-reviewers. An editor can stop a paper from going to peer review. He can pull the paper out of the process after receiving even a hint of criticism, or he can ignore much of the criticism and push the paper forward.

The decision of the editor is typically final and without appeal. The recommendation of a reviewer is always only a recommendation, it can always be ignored by the editor.

So, really, if you feel that it is important that your identity be hidden, it should be hidden from the editor.

But anyway, I’m glad that turning the tables on the reviewers (i.e., making them non-anonymous) seems like a good idea to you, too, so let’s work on that, it looks like a good start,

Again, thanks for replying.

Please report issues using either github or preferably, Google code:

http://code.google.com/p/lemurbitmapindex/issues/list

Be sure to specify which compiler you are using.

headers/ewahutil.h:21:23: error: conflicting declaration ‘typedef long unsigned int uint32_t’
/usr/include/stdint.h:52:23: error: ‘uint32_t’ has a previous declaration as ‘typedef unsigned int uint32_t’
make: *** [unit] Error 1

can yout tell me how to fix it.

I’ve implemented memoization. The result is poor. Please see my implementation:

https://github.com/lemire/Code-used-on-Daniel-Lemire-s-blog/blob/master/2012/05/21/testbitmap.java

Reviewers “suspect” who the author is, “google” his/her prior work, fail to check derivations based on the authors’ reputation, and spend their otherwise precious time on inconsequential aspects – focus on the paper given to you for review, and do not concern yourself with the author and his affiliation. Look, if you don’t have time to do a proper review, THEN DON’T DO IT

The person taking the decision to accept or reject the paper, the editor, knows your reputation and your affiliation. So, sadly, there is no way to take this out of the equation. In fact, in a double-blind system, the editor has even more power.

To properly review a paper, I need to compare it with the state-of-the-art and previous work. Typically, this means reviewing (even unknowingly) what the author worked on previously. So, you cannot just focus on the paper as a separate entity.

(…) fairer would be to disclose the identity/affiliation of the reviewers to the author, so that he/she could have a better idea about how competent they really are to judge his/her work. How comfortable would reviewers feel in that situation?

I have been advocating precisely this scheme for years. If you are going to criticize someone, you should be able to do it openly, in front of witnesses, not under the cover of anonymity.

But it works both ways… if you are going to ask me to review your work, through a journal or otherwise, please let me know who you are.

Once everything is public, the focus will naturally be on the science. If people unfairly review your work (because you lack a prestigious affiliation) then you’ll be able to point to the evidence, as it will all be out there. It also takes power away from the editors, and it puts it back in the community at large.