Показать все сообщения Спрятать все сообщенияJust curious did anyone managed to pass the time limit
using binary search + rolling hash or the test data are
specially designed to be solved in time only by an efficient suffix tree/array implementation? I tried an O(N.log N) suffix array (although finding the length of a match was O(N.log N.log N)) and it was too slow, although it ran in about 1.3s on my machine. Test 10 was very hard for binary+hash solutions,but
double or triple hash would pass it. These solutions are correct, because the test set was really hard. The solutions of the authors are based on suffix trees (Nikita Rybak, Pascal) and suffix arrays (Ilya Grebnov, Java) and pass the TL easily. Thank you Dmitry!
It seems that pretty soon Ukkonen and/or efficient suffix array implementations
will become a must in real time contests.
I would like to congratulate the autors for the excellent problem set! Single hash-based solution got AC too I got AC with optimized single hash-based solution too. But if
I wrote double hash based solution I wouldn't have +9. I got TLE on #28, with single hash solution. The simplest solution with single hash gets AC in 0.468 Some statistics:
0. Suffix Automaton get AC in C++ 0.125 time and 22 320 KB Memory, author is [SPbSU ITMO] WiNGeR
1. Suffix Automaton get AC in C++ 0.14c with 22220КБ Memory, author is Burunduk1
3. Suffix Tree gets AC in 0.14 time and 23 888 KB Memory, author is [SPbSU ITMO] WiNGeR
4. Suffix Array with O(N) implementation get AC in C++ 0.343c with 5756 КБ Memory, author is me
5. Single Hash get AC in C++ 0.39c with 4024 КБ Memory, author is me
6. Single Hash get AC in Pascal 0.5c with 6191 КБ Memory, author is me
7. Suffix Tree get AC in Pascal 0.656c with 16416 КБ Memory, author is Kit(Vologda SPU)
8. Suffix Array with O(NLogN) implementation get AC in C++ 0.765c with 2756КБ Memory, author is me
9. Suffix Array with O(NLogN) implementation get AC in Java 1.406c with 5503КБ Memory, author is me
Edited by author 22.12.2006 15:35
Edited by author 25.12.2006 11:24
Edited by author 11.02.2007 13:44
Edited by author 11.02.2007 13:45
Edited by author 11.02.2007 13:45 I write Suffix Tree(Ukkonen Algorithm).
AC - 1 sec.
But I used 30mb memory. I tried to solve this problem using Suffix Tree (ukkonen95) and got MLE on test #9, what trick did you use to reduce the memory?
Then I saw this discussion about Suffix Array, I tried Larsson-Sadakane algorithm and got WA on test #8. Then I tried Karkkainen-Sanders algorithm and got TLE on test #9.
I haven't try to use Suffix Automaton yet.
This problem is really hard. Anyone can give hints about reducing memory for the Suffix Tree? I tried to solve this problem using Suffix Tree (ukkonen95) and got MLE on test #9, what trick did you use to reduce the memory? No tricks, I have almost double reserve in memory. But I use hash working with edges, may be that's the matter. Test #8 is extremally small, so you can find similiar manually. #9 is just the first large test. Good luck! No tricks, I have almost double reserve in memory. But I use hash working with edges, may be that's the matter. Yes, I got it Accepted at last :) The trick is to use your own hash implementation rather than hash_map from STL! Did anybody solve this problem using STL? I just got AC with Suffix Array + LCP in 0.3xx secs.
The implementation is cleaner than Suffix Tree and requires less space yet still run in O( N ) time.
I'm curious with Burunduk1's Suffix Automaton, is there any online resources available to learn about Suffix Automaton? I really want to solve this problem using Suffix Automaton :) >This problem is really hard. Anyone can give hints about >reducing memory for the Suffix Tree?
I also had MLE#9, after that I saved at node of suffix tree only used links to child And got AC.
It reduced memory, but I my prog eat 30mb. Those who solved it using Suffix Trees:
Did you built 2 Suffix Trees or just 1?
What I did is to create 1 Suffix Tree with string: s1 + "$" + s2 + "#" where s1 and s2 is the first and second string in the input. After I built the Suffix Tree, I traversed it and find the deepest node which has child "$" and "#".
Or is it better to create 2 Suffix Trees and compare the two trees?
Or there is a way to build a Suffix Tree only for the first string and try to find the LCS with the second string using that Tree.
Did you get AC using the first algorithm? second? or the last? Thanks I use hashing but my algorithm is O(n(logn)^2), is it a better way?
Here is my algorithm:
m = the result
l = length(m)
- Do a binary search on l
Check whether there is a common substring length l:
+ Calculate and sort hash values of string 1 (into array f)
+ For each value in string 2, search it in f, check whether
there is a match! What is double hash???
My single hash O(N*logN) prog. got AC within 0.671 sec. and 6,... MB memory. What I did is to create 1 Suffix Tree with string: s1 + "$" + s2 + "#" where s1 and s2 is the first and second string in the input. After I built the Suffix Tree, I traversed it and find the deepest node which has child "$" and "#". this is exactly what iam trying to do but i got stack overflow in test case 9, obviously the stach overflows when i traverse the tree wishing to find the deepest node which has the two terminator characters as children (or in the subtree below). I guess the tree is too deep, so how could u solve this part? Don't traverse the tree using recursion ;)
Try using explicit array[100000] Don't traverse the tree using recursion ;)
Try using explicit array[100000] Do u mean to use that array to bottom up traverse the tree? or u mean something else? Don't traverse the tree using recursion ;)
Try using explicit array[100000] Do u mean to use that array to bottom up traverse the tree? or u mean something else? A recursion is using stack implicitly (stack memory) and it has limitation on the depth of the recursion for some compiler (or maybe by the stack memory configuration of the compiler). So, instead of using stack memory we can use an explicit stack array to simulate the same behavior. However, the code gets uglier and harder to read :p Update [SPbSU ITMO] WiNGeR 9 фев 2007 20:51 Update [SPbSU ITMO] WiNGeR 9 фев 2007 20:57 Suffix Tree gets AC in 0.14 time and 23 888 KB
Suffix Automation gets AC in 0.125 time and 22 320 KB
autor is me
P.S. Who can tell me, how to implement Suffix Array in O(n)? Re: Update Ilya Grebnov[Ivanovo SPU] 11 фев 2007 13:41 Suffix array can be constructed in O(n) time by Karkkainen-Sanders algorithm. Hi WiNGeR, You might be interested to see this demo about creating Suffix Array in linear time by Juha Karkkainen and Peter Sanders: http://felix-halim.net/pg/suffix-array/index.php Do you have any reference on how to build Suffix Automaton? Which one is simpler? Suffix Tree or Automaton? Thanks i relalize double hashing, but i get tle on test 9.
Can anybody get me hash-functions used in solution on this problem? I can't use single hash to solve the problem...
Would you please tell me how to design the hash?
And what is the 64bit hashing meaning? As far as I am concerned, Suffix Automata are the best tool for solving this problem. At least, it holds for me. I got AC three hours after I started reading "Automata for Matching Patterns" ( http://www-igm.univ-mlv.fr/~mac/REC/B4.html ). Earlier I spent much more time on Suffix Trees and Arrays - and I failed to solve this problem by their means. Would be grateful for any hints concerning Trees and Arrays (just curious). I used the standard ukkonen implementation. However, I use sibling lists for edges representation (watch Wikipedia "Suffix tree"), and I make leaves of tree explicit, that means that in the worst case you have 2*N nodes totally, N of them are leaves, where N is the total size of the text. This works fine.
Also, for the traversal of the tree, I used simple recursion. I tried to use stack modelling, as someone proposed in the posts earlier, but failed with TLE on test 9. In order not to get stack overflow set up the size of the stack about 10 megabytes (watch FAQ in order to learn, how to do it). nlogn georgi_georgiev 31 авг 2009 13:14 I just did it with hash and my own hash table for 0.265 sec and 7mb memory. I used the standard ukkonen implementation. However, I use sibling lists for edges representation (watch Wikipedia "Suffix tree"), and I make leaves of tree explicit, that means that in the worst case you have 2*N nodes totally, N of them are leaves, where N is the total size of the text. This works fine.
Also, for the traversal of the tree, I used simple recursion. I tried to use stack modelling, as someone proposed in the posts earlier, but failed with TLE on test 9. In order not to get stack overflow set up the size of the stack about 10 megabytes (watch FAQ in order to learn, how to do it). Totaly agree. Sibling lists save memory well, and work pretty fast. I also needed to increase stack size. http://acm.timus.ru/status.aspx?author=48362&from=3761647&count=1
Edited by author 27.10.2011 02:46
Edited by author 27.10.2011 02:46 Hi!
I have tried so many interpretations of the main idea for finding suffix array in O(NlogN), but all of them (except 1, because I optimize it) ran for more than 2 second .
Would you share the secret, I mean is there something special you do? There is no any secret. Try to find original article "Faster Suffix sorting" by N.Jasper Larson and Kunihico Sadakane. Also very useful article for this problem is "Two Space Saving Tricks for Linear Time LCP array Computation" by Giovanni Manzini. Just got 0.96s and 28M memory with suffix automaton. How do i make it fast? String hashes + binary search: 0.468 s
String hashes + hashset: 0.234 s Suffix automation AC C++ MSV2013 0.093 sec 22 MB, yeah Actually, not any hash is suitable here. Why not suffix automaton?
It is much easier to implement and, of course, gets AC =) |
