nocriz's blog

Technocup 2020 — Elimination Round 3 + Codeforces Round 602: Editorial
By nocriz, 5 years ago, In English
1262A - Math Problem

Writer: RDDCCD

Tutorial
1262B - Box

Writer: RDDCCD

Tutorial
1261A - Messy

Writer: RDDCCD

Tutorial
1261B1 - Optimal Subsequences (Easy Version)

Writer: MikeMirzayanov

Tutorial
1261B2 - Optimal Subsequences (Hard Version)

Writer: MikeMirzayanov

Tutorial
1261C - Arson In Berland Forest

Writers: BledDest, adedalic

Tutorial
1261D1 - Wrong Answer on test 233 (Easy Version)

Writer: RDDCCD

Tutorial
1261D2 - Wrong Answer on test 233 (Hard Version)

Writer: RDDCCD

Tutorial
1261E - Not Same

Writers: nocriz, RDDCCD

Tutorial
1261F - Xor-Set

Writer: nocriz

Tutorial
  • Vote: I like it
  • +113
  • Vote: I do not like it

Comments Comments (50)
| Write comment?
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

Auto comment: topic has been updated by nocriz (previous revision, new revision, compare).

→ Reply
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

The tutorial for 1261B — Optimal Subsequences is not ready, but I feel that it is better I post the tutorial soon! That problem is not by me and I hope the tutorial will be available soon.

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it +11 Vote: I do not like it

    I can just post the brief solution:

    • The optimal subsequence for a given $$$K$$$ contains the largest $$$K$$$ values.
    • If the indices with the smallest of these values can be chosen in multiple ways, we want to choose smaller indices.
    • That means we want to sort the sequence of all $$$(A_i, -i)$$$. We get the answer to a query by taking the last $$$K$$$ pairs, sorting the indices in them and taking the $$$id$$$-th of these indices.
    • Easy offline solution: sort queries by $$$K$$$, add pairs one by one, build a sorted array of indices in them either using a treap or some sqrt structure.
    → Reply
»
5 years ago, # |
  Vote: I like it +3 Vote: I do not like it

Qualifying round for Technocup (competitions for CIS schoolchildren), but tutorial in English only. hmm

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it +30 Vote: I do not like it

    Editorial will be available in Russian soon.

    → Reply
»
5 years ago, # |
  Vote: I like it -19 Vote: I do not like it

obviously, its' easy and similar words are not appropriate for editorials. I understand that once you know the solution it obvious. There are thousands of people who were didn't found it obvious.

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Which part exactly do you need us to explain? I'll be glad to answer your question.

    → Reply
    • »
      »
      »
      5 years ago, # ^ |
        Vote: I like it -22 Vote: I do not like it

      A-Messy It's easy to construct a valid bracket sequence, I am puzzled if we can construct any valid sequence? If so probably we put all left brackets to left and right to right side, but I think we need to construct K outer brackets. maybe code solution example would help

      → Reply
      • »
        »
        »
        »
        5 years ago, # ^ |
          Vote: I like it +14 Vote: I do not like it

        65626236 I would recommend you to see tourist's solution.

        → Reply
      • »
        »
        »
        »
        5 years ago, # ^ |
        ← Rev. 2   Vote: I like it +10 Vote: I do not like it

        I think "valid" is supposed to mean "with exactly K outer brackets". The example given in the tutorial is ()()()()((((())))) for say K=5; so you just put K-1 empty brackets first, then all the left brackets, and finally all the right brackets. This will always give you a valid bracket sequence with K outer brackets.

        → Reply
»
5 years ago, # |
← Rev. 9   Vote: I like it +15 Vote: I do not like it

My $$$(n + q) \ log \ n$$$ solution for 1261B2 - Optimal Subsequences (Hard Version)

Greedy
Binary Indexed Tree(BIT)

Code : 65736937

→ Reply
»
5 years ago, # |
← Rev. 3   Vote: I like it 0 Vote: I do not like it

Why the following statement in Problem F is true? Shouldn't it be $$$4 \times n^2$$$?

We can prove that the number of both "real" and "auxiliary" segments of any size is not greater than 4 x n.

→ Reply
  • »
    »
    5 years ago, # ^ |
    ← Rev. 2   Vote: I like it -10 Vote: I do not like it

    Consider a access to [l,r] on the segment tree, we would call the segments we accessed and found that the segment is completely in [l,r] real segments, and all segments we visited are auxiliary segments. Since we tried visiting n intervals (All intervals in A or B), and for each visit the segments (nodes) we touched of each depth is not greater than 4, the number of both "real" and "auxiliary" segments of any size is not greater than 4 x n is proved.

    This is the same with the proof of the complexity of the segment tree, I found this on internet only 4 nodes are processed in a level.

    Sorry, but I have no idea why it could possibly be $$$4n^2$$$.

    Check out my solution for more information:65749995

    → Reply
»
5 years ago, # |
← Rev. 2   Vote: I like it 0 Vote: I do not like it

For anyone looking for the solution of 1261B2 - Optimal Subsequences (Hard Version) in Java, you can refer to my solution. I used the same approach as mentioned in the tutorial and used Segment Tree to find the kth smallest element in the set. 65750497

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Btw no need to binary search on seg tree queries, you can use this trick, (ctrl-f for find_kth)

    → Reply
    • »
      »
      »
      5 years ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      Thanks a lot, exactly what I was looking for. It decreased the time complexity from log(n)^2 to log(n). Thanks a lot.

      → Reply
»
5 years ago, # |
  Vote: I like it +1 Vote: I do not like it

Thank you so much for the contest and editorials!

I'm just wondering if anyone has a solution which runs under $$$O(n^2)$$$ for Div1A/Div2C (Messy)? Or alternatively, is there a way to show that it is not possible to do this with better time complexity?

Thanks!

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    I was wondering the same.

    After spending 5 mins thinking about a solution, I noticed n^2 can also pass but it looked like it can be solved in better complexity too.

    → Reply
  • »
    »
    5 years ago, # ^ |
    ← Rev. 3   Vote: I like it 0 Vote: I do not like it

    There are several ways to implement a solution that runs under $$$O(n^2)$$$.

    Both operations explained in the editorial can be implemented in $$$O(log(n))$$$ using treap: code

    Also, you can notice that in this problem, reversing a range is only going to change the value of two positions (start and end of the range). So you can implement the reverse operation with a simple segment tree (by updating both ends of the range only). Finding the first position to the right equal to a value can be done in more than one way using the same segment tree. For example:

    • Binary search over prefixes starting in each position in $$$O(log^2(n))$$$: code.
    • Descending over the nodes of the segment tree, achieving $$$O(log(n))$$$: code.

    I didn't explain every detail in case someone wants to think about them. I'm not sure whether a $$$O(n)$$$ solution exists. Feel free to ask anything :)

    → Reply
  • »
    »
    5 years ago, # ^ |
    ← Rev. 4   Vote: I like it +11 Vote: I do not like it

    Actually it can be done in $$$O(n)$$$. Here is my code: 65767515

    The main idea is to create pointers first_opened and first_closed for first ( and ) in s and just update them. The solution is similar to the solution in the editorial. But every time when $$$s[i] \neq t[i]$$$, look at the substring s[i..max(first_opened, first_closed)]. It is either (..() or )..)(. Reverse can be done in $$$O(1)$$$ and after that you have to fix pointers, it is $$$O(n)$$$ total (details in the code).

    Also, I'm pretty sure that there is no $$$o(n)$$$ solution :)

    → Reply
    • »
      »
      »
      5 years ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      My solution works in O(n) time which I think is a bit different than yours. 67442807

      → Reply
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

I saw solutions for B1, B2 using Segment Tree and BIT. How can we find kth smallest element in set using those 2 data structures? I solved it using ordered_set (PBDS in GNU).

Also, is insertion operation also supported with this trick (with the use of BIT/SegTree)?

→ Reply
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

You can do the hard version of optimal subsequences with online queries using sortings and a mergesort tree. Here is my solution: https://codeforces.net/contest/1262/submission/65675557

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Also you can compute persistent segment tree for each prefix that will store for each value count of it. And you will just go down in this tree in O(log n) to find min prefix with sum >= k.

    → Reply
»
5 years ago, # |
← Rev. 3   Vote: I like it 0 Vote: I do not like it

The issue is solved.

→ Reply
»
5 years ago, # |
← Rev. 2   Vote: I like it 0 Vote: I do not like it

Can somebody please explain why we should multiply the answer by k^(n-t) in div1D/div2F i think the n-t other answers have fewer possibilities

→ Reply
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

Sorry, but I have tried hard but still can't understand the given solution of 1261E ,I wonder what's the meaning of operations for {4,1} and the later stuff? Looking for your reply (:

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    I'm sorry if the editorial is hard to understand. {4,1} means that we currently iterated through two elements in A, namely 4 and 1. If the next element in A is 1, then there are three numbers we handled, {4,1,1}. Be careful not to mix it up with the "compressed set".

    → Reply
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

I solved problem E using the concept entropy, you can see it here.

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Can you give a brief explanation of your solution? How have you used entropy here?

    → Reply
    • »
      »
      »
      15 months ago, # ^ |
        Vote: I like it 0 Vote: I do not like it

      First, I don't have a proof for my solution, but I can explain the process to let you understand what I'm thinking. First, change the original problem to this: Make $$$n$$$ subsets of $$${1,...,n+1}$$$,call them $$$T_1,...,T_n$$$. And, these should hold: $$$|T_s|=a_s$$$, for any $$$i\neq j\in [1,n+1]$$$, $$${k|i\in T_k}\neq{k|j\in T_k}$$$. This means, we should use $$$n$$$ sets to tell $$$n+1$$$ elements apart. And now, there comes the entropy. If you have some sets $$$T_1,...,T_s$$$, then you use these $$$s$$$ sets to make the $$$n+1$$$ elements some classes $$$C_1,...,C_q$$$, and the entropy of this state is:

      $$$S(T_1,...,T_s):=-\sum_{t=1}^q[|C_q|/(n+1)]\log[|C_q|/(n+1)]$$$

      and the goal is to make $S(T_1,...,T_n)=(n+1)\log[n+1]$. Until now, there is nothing unproved. To achieve the goal, I choose greedy. Say, I already have $$$T_1,...,T_s$$$, how do I choose $$$T_{s+1}$$$, the way I use is to maximize $$$S(T_1,...,T_s,T_{s+1})$$$ without change the choice of $$$T_1,...,T_s$$$.

      → Reply
      • »
        »
        »
        »
        15 months ago, # ^ |
        ← Rev. 2   Vote: I like it 0 Vote: I do not like it

        I found that I was stupid. The proof for this is really easy, and, the proof will lead to a much easier way to solve the problem. I didn't read the tutorial, but I guess the solution tutorial is the "much easier way". Because for each $$$s$$$, there is at least one way to choose the $$$T_s$$$, so the number of classes will add one.

        → Reply
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

Problem : Arson in Berland Forest

How to do "add value on a rectangle" (using prefix sums)" ?

→ Reply
»
5 years ago, # |
← Rev. 3   Vote: I like it 0 Vote: I do not like it

.

→ Reply
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

What if we want to minimize number of operations in 1261A?

→ Reply
»
5 years ago, # |
  Vote: I like it 0 Vote: I do not like it

How to solve 1261 C if we should also minimize the initial number of burnt trees for a particular maximum value?

→ Reply
»
5 years ago, # |
← Rev. 3   Vote: I like it 0 Vote: I do not like it

1261C (Arson in Berland Forest) is doable in $$$O(nm)$$$ time by finding the largest square with lower-right corner at each cell, and then applying the skyline algorithm twice (once vertically, once horizontally on the result of the vertical). Normally the skyline problem is solvable only in $$$O(n\log n)$$$ time, however in this case we don't have any "nested" segments (i.e. a pair of segments s.t. one is higher the other and also contained within it), so I was able to solve it using a deque rather than a priority queue (as would normally be used to solve skyline). This allows us to generate a table M s.t. M[i][j] is the size of the largest square which contains the cell (i, j). Now just find the smallest positive element of M and that's the answer. (From here, finding a valid set of initial points is trivial)

My code here: 74775877

→ Reply
  • »
    »
    5 years ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    emorgan Can you explain how have you used skyline to create M[i,j]? Hard to understand from your code.

    → Reply
    • »
      »
      »
      5 years ago, # ^ |
      ← Rev. 2   Vote: I like it 0 Vote: I do not like it

      First, I build a table sz where sz[i][j] is the size of the largest square with lower-right corner at cell (i, j). This is a classical problem solved by dynamic programming.

      Then, I consider each row one at a time. For each index j, I would like to update the values of sz to the left of j, in a way that makes it such that the value in each cell corresponds to the size of the largest square containing that point, with lower-right corner somewhere in that row. To do this, I simply consider all such squares (i.e. all values in that row), and perform range max updates on that specific row on the interval covered by each square. The skyline algorithm actually allows us to do this in $$$O(n)$$$ time. The way it works is quite complicated, and I don't think I know how to explain it at a basic level.

      Then, we just do the same thing on columns instead of rows, and sz now holds the values of the desired array M which i mentioned in my previous comment.

      → Reply
      • »
        »
        »
        »
        5 years ago, # ^ |
        ← Rev. 3   Vote: I like it 0 Vote: I do not like it

        Thanks emorgan! Your this line makes sense to me — "update the values of sz to the left of j, in a way that makes it such that the value in each cell corresponds to the size of the largest square containing that point, with lower-right corner somewhere in that row". Some print statements in your code helped me to understand the while loops in skyline algo. As per my understanding, use of the first, second while loop is as per image1, image2 respectively.
        image 1 — https://www.dropbox.com/s/ixjh6pbje3tbv0j/pic1.jpg?dl=0

        image2 — https://www.dropbox.com/s/571zalnyq5mzs8m/pic2.jpg?dl=0

        But why are we doing skyline columnwise as well? After rowwise operation we already have max square which contains (i,j).

        → Reply
        • »
          »
          »
          »
          »
          5 years ago, # ^ |
            Vote: I like it 0 Vote: I do not like it

          After the row-wise operation, sz[i][j] stores the size of the largest square which contains (i, j), and also has lower-right corner in that same row. However, we need to consider squares from lower rows as well. So we perform a second column-wise pass to propagate the values from lower rows upwards. The end result is that each cell storing value $$$s$$$ updates an $$$s$$$ by $$$s$$$ square up and to the left, rather than just to the left along the row.

          → Reply
»
4 years ago, # |
  Vote: I like it 0 Vote: I do not like it

D2 can be solved very easily using ordered_set (for c++) https://codeforces.net/contest/1227/submission/88584680

→ Reply
»
4 years ago, # |
  Vote: I like it +3 Vote: I do not like it

I did Arson in Berland Forest by using (inventing ^_^) 2-dimensional sweepline algorithm. Have a look : 92480981

→ Reply
»
4 years ago, # |
  Vote: I like it 0 Vote: I do not like it

In DIV2E/DIV1 C, the wording of this statement could have been better You are sure that all burnt trees are shown on the map. All the trees outside the map are undamaged. " are undamaged " should have been replaced with " should be undamaged "

→ Reply
»
4 years ago, # |
  Vote: I like it 0 Vote: I do not like it

I tried to solve problem D2 of Div-2 using segment tree, where I am just assigning new indexes discovered in queries(sorted) to 1 and finding the pos-th 1. But it is giving WA on test 10. Can anyone please help me in finding what I did wrong. Thanks.

https://codeforces.net/contest/1262/submission/110093210

→ Reply