Блог пользователя maroonrk

Автор maroonrk, история, 5 месяцев назад, По-английски

We will hold estie Programming Contest 2023 (AtCoder Regular Contest 169).

The point values will be 400-500-600-700-900-900.

We are looking forward to your participation!

  • Проголосовать: нравится
  • +132
  • Проголосовать: не нравится

»
5 месяцев назад, # |
  Проголосовать: нравится +30 Проголосовать: не нравится

How to solve A? 🤡

  • »
    »
    5 месяцев назад, # ^ |
      Проголосовать: нравится 0 Проголосовать: не нравится

    Consider the "contribution on the answer" by each depth of the tree. This "contribution" increases overwhelmingly as the depth increases. So, for each depth, from leaf to root, consider the sum of the values on that height. If this sum is nonzero, this sum will overwhelm the value in finite time. If this sum is zero, this sum does not affect anything. Run a greedy based on this idea.

  • »
    »
    5 месяцев назад, # ^ |
    Rev. 2   Проголосовать: нравится 0 Проголосовать: не нравится

    make graph out of P, edges $$$a[i] -> i$$$, $$$2 <= i < N$$$, now see that the nodes at more distance from 1 will dominate, so for all nodes at d from 1 find the sum of $$$a[i]$$$ of all such nodes, the largest d for which sum is non zero will dominate.

»
5 месяцев назад, # |
  Проголосовать: нравится 0 Проголосовать: не нравится

I haven't solve anything. I don't understand it's not my day, or the contest was quite difficult ?

»
5 месяцев назад, # |
  Проголосовать: нравится 0 Проголосовать: не нравится

How to solve B!?

»
5 месяцев назад, # |
  Проголосовать: нравится +21 Проголосовать: не нравится

Trying to solve C == Communicating with aliens

  • »
    »
    5 месяцев назад, # ^ |
      Проголосовать: нравится 0 Проголосовать: не нравится

    Actually for me, it felt quite different (not necessarily easy, but still in a good way) compared to C's in previous ARCs. If you can come up with a trivial $$$\mathcal{O}(N^3)$$$, the rest should be simply optimizing it to $$$\mathcal{O}(N^2)$$$.

    • »
      »
      »
      5 месяцев назад, # ^ |
        Проголосовать: нравится 0 Проголосовать: не нравится

      But I thought $$$\mathcal O(n^2)$$$ is completely different from $$$\mathcal O(n^3)$$$ solution.

      My method is:$$$g[i][j]=$$$ ways to determine $$$a_1\cdots a_i$$$ and $$$a_i\neq j$$$.

      • »
        »
        »
        »
        5 месяцев назад, # ^ |
        Rev. 2   Проголосовать: нравится +6 Проголосовать: не нравится

        My method was, $$$dp[i][j]$$$ is number of prefix up to current element where $$$i$$$ comes last, repeated $$$j$$$ times. (transition is done inplace) This trivially leads to a $$$\mathcal{O}(N^3)$$$ DP. However, in here the greatest bottleneck in DP is simply shifting the DP table by one position which takes $$$\mathcal{O}(N^2)$$$ alone in the $$$a_i=-1$$$ case. So I used deques for the DP table, making it $$$\mathcal{O}(1)$$$ per row for shifting. Clearing a row is also tricky but it can be done in amortized $$$\mathcal{O}(1)$$$ using rollback ideas. This led to an $$$\mathcal{O}(N^2)$$$ solution for me.

        Implementation link

      • »
        »
        »
        »
        5 месяцев назад, # ^ |
          Проголосовать: нравится +25 Проголосовать: не нравится

        My dp method is same as yours, but I solve it with inclusion-exclusion method in $$$O(n^2)$$$. submission

»
5 месяцев назад, # |
  Проголосовать: нравится +22 Проголосовать: не нравится

felt like B < C < A

»
5 месяцев назад, # |
  Проголосовать: нравится 0 Проголосовать: не нравится

For problem C, I use dp[i][j] to denote the number of ways, under the condition that, all the -1 in the first i positions have benn determined, and the last element is j.

dp[i][j] = sum of dp[i-x][y], where x=1,2,..,min(j, f(i,j)), and y=1,2,...,j-1,j+1,...,n, and function f(i,j) denotes the maximum number of j that we can have in a row from the i-th position to the left.

Note that for some fixed x, sum of dp[i-x][y] for y=1,2,..,j-1,j+1,..,n, can be computed by dpt[i-x]-dp[i-x][y], where dpt[i-x]=dp[i-x][1]+dp[i-x][2]+...+dp[i-x][n]. Next, sum of dpt[i-x]-dp[i-x][y] for x=1,2,..,min(j, f(i,j)), can be computed by prefix technique again.

»
5 месяцев назад, # |
Rev. 2   Проголосовать: нравится 0 Проголосовать: не нравится

can somebody explain to me why we need to find the depth? if the value gets replaced so A[P[i]] becomes A[P[i]] + A[i] the index still doesnt change as in P[i] doesnt become something else, only A changes, in other words only the same couple of numbers actually change A[1], can somebody tell me where did i go wrong? Problem A btw

  • »
    »
    5 месяцев назад, # ^ |
      Проголосовать: нравится 0 Проголосовать: не нравится

    First, the modification relationship can be built as a tree. The value of a node will only be modified by its children. So the deeper the point the higher the priority. Comparisons are made according to depth, and if the sum of the nodes in the current layer is not 0, the positive and negative outputs can be judged, otherwise the next layer up is compared until the point 1 is reached.

    I'm not good at English. I hope this helps :)

  • »
    »
    5 месяцев назад, # ^ |
      Проголосовать: нравится 0 Проголосовать: не нравится

    My thinking was the same as yours at first, but then I realized that different nodes have different effects on A[1]. This is because some nodes that affect A[1] are themselves affected by other nodes.

»
5 месяцев назад, # |
  Проголосовать: нравится 0 Проголосовать: не нравится

does someone have a simpler solution for D?