AtCoder Failure List
MLE
Construct a graph from a given H x W = 10^6 string to TLE or MLE PAST5H. TLE
If we take the list of "next vertices to visit" of the whole search and turn it into a TLE set, then AC ABC184E. Forgot to put "visited" on Dijkstra implementation PAST4J. Preprocessing to speed up, two things to preprocess but only one of them is done, TLE, misunderstanding the computation order and making an ad hoc constant times faster PAST4K. One max was inside the loop to calculate the maximum score (TLE) ABC175D. Creating a power table in preprocessing: I was doing a ** x and the digits exploded before I got to the remainder / TLE to pow(a, x, MOD) / This is logarithmic order, so when creating the table I should have made it constant order by reusing the previous values ARC106D Constant order, but the constant is 10^9 ABC164D. Depth-first search is visit(pos) when it should be visit(next) (you should have noticed this if you tested on hand).
Implemented depth-first search with recursive function calls, but PyPy is slow to call functions, and the number of times is strictly on the order of 10^6, rewrote to a while loop PAST5H. Join strings in a loop, TLE, append to a list, and join at the end PAST5H. For repeated multiplication of small matrices, using Numpy is less beneficial because of the large overhead ABC189E AC ABC190E if you use Dijkstra to change to TLE and BFS even though all sides cost 1. TLE by creating a table of 2^16 to remember "what numbers have already appeared" in the digit DP, and only "how many have already appeared" since all kinds of numbers always appear in the digit DP less ABC194F. RE
N allocates 10^5 interval DP, 10^10 memory
I should have realized that "we'll never get it done in time with DP." AGC048B AC: Cumulative sum to collapse the innermost loop and AC ABC179D. AOJ and my Python are not working, and AtCoder's code test says 10^6 is OK, so it probably depends on the compile options of the processor.
WA
That I assumed that there was only one connected component that had a size greater than 2. In fact, there is more than one ARC107C Corner case where N=1 in a question about the relationship between multiple things ARC106C I'm setting INF to 10**10 and there are values that exceed AGC044A. I compressed the coordinates but accessed them with the original coordinates PAST2N. Negative edges in the minimum cost flow library that should not contain negative edges PAST3O. Don't put a negative side in the Dijkstra Law either.
Continue when the pointer is incremented even though it is a for statement ABC178F. I wanted to know if some of the vertices were reachable, but I checked whether the INF was included in the Dijkstra result to see if "all vertices were reachable" ABC190E. The initial value of right in a binary search satisfies the condition ABC192D When you get to WA and fix the bug and submit a second time, you do a "fix the bug, lightly speed up" and then put the bug in that speed up. They think the bug isn't fixed and they get confused ABC192F Mathematically there is an O(1) solution, but there is a floating-point number error, I stuck with the O(1) solution, but O(logN) is the correct solution ABC191D with integer and binary search. ---
This page is auto-translated from /nishio/AtCoder失敗リスト using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.