### ilyakrasnovv's blog

By ilyakrasnovv, 2 years ago,

Hello, codeforces!

We deeply apologize for the weak pretests in the problem 1642C - Great Sequence. Anyway, we believe that you liked at least one problem from the contest :)

Thanks to Mangooste for the problem 1641E - Special Positions!

Solution: 147513897

Solution: 147513934

Solution: 147513962

Vector solution: 147513974

$\mathcal{O}(2n^2)$ insertions solution: 147514019

$\mathcal{O}(n^2)$ insertions solution: 147514028

Solution: 147514055

Set solution: 147514064

Solution: 147514090

Bitset solution: 147514108

Solution: 147514167

Solution: 147662463

• +363

| Write comment?
 » 2 years ago, # |   +4 It doesn't allow to show the solutions
•  » » 2 years ago, # ^ |   +8 We'll fix it in several minutes, sorry
•  » » » 2 years ago, # ^ |   +7 Done
 » 2 years ago, # | ← Rev. 4 →   +101 I guess most participants would consider the bitset solution to D cheese, but I'd like to point out another way to optimize it.Naturally, the bitset solution would be like: sort the arrays by $w$; compress the numbers, for each number store the bitset of the arrays it appears in; iterate over the first array, OR the bitsets of the numbers in it, then take _Find_first() of its complement. The only issue is that you can't store all bitsets, it takes $O(n^2m)$ memory. The authors overcome this by solving the problem separately for numbers that appear less than $T$ times and more by choosing $T$ carefully.But there is a common problem, where we encounter the same issue. (As memed as it is) It's called "count the number of pairs of reachable vertices in DAG" or something along these lines. How do we deal with it there? Well, we process vertices in batches of $64$.Let's do the same. Process arrays in batches of $64$. Consider the first batch of arrays from $1$ to $64$. For each number, build a bitset of size $64$ (also known as an unsigned long long) with arrays from the batch it appears in. This is done in $64m$ operations. Then iterate over all arrays and do the same OR thing as in the initial solution (find first becomes __builtin_ctzll). With this, we only consider pairing each array with some array from the batch. We have to process $\frac{n}{64}$ batches, and it takes $O(nm)$ to process each one. So we get the solution with the same complexity but with just $O(nm)$ memory.Code: 147454534
•  » » 2 years ago, # ^ |   +5 Thanks for the solution! But I have a question: "Then iterate over all arrays and do the same OR thing as in the initial solution (find first becomes __builtin_ctzll). With this, we only consider pairing each array with some array from the batch."But in your submission in line 124 we have this: if (pos < n) ans = min(ans, w[i] + w[pos]);Shouldn't it be if (pos < r) ans = min(ans, w[i] + w[pos]);?I changed $n$ to $r$ and it's also AC.
•  » » » 2 years ago, # ^ |   +3 That line only affects the behavior on the last batch. On any other batch you will never get pos outside l..r-1 range because of the check that the mask is not full.
•  » » » » 2 years ago, # ^ |   0 oh, got it, thanks!
 » 2 years ago, # |   +63 I think it's worth mentioning that 1641C - Анонимность наше все can be solved with a DSU, which quite a few people did. The way I thought of it is that we consider the prefix sum array of the number of ill people (so for example, a segment of healthy people corresponds to a segment in this array where the value is constant). The DSU stores collections of indices which are known to have the same value (these components are always segments) as well as a pointer to the next position that is known to be in a different component. The time complexity is $O((n+q)\alpha^{-1}(n))$ which is better than the editorial. (Maybe it could be improved to $O(n+q)$ if we account for the simple structure of the components?)
•  » » 2 years ago, # ^ |   +3 vector least_enemy; In your submission using DSU, what does the least_enemy vector represent ? Also your written code is really clean. very easy to read. Thank you.
•  » » » 2 years ago, # ^ |   0 If x is a root, then the component of least_enemy[x] is the leftmost component after the component of x which is known to be disjoint from the component of x.
 » 2 years ago, # |   0 Can someone help me find error/mistake in my submission for A ? Thanks ! Code
•  » » 2 years ago, # ^ |   0 The ans must be an integer, so don't use double for precision loss.
 » 2 years ago, # |   0 For problem div 2E, there's this statement find the next one and do the same thing until we find the first index greater than r I don't quite get it. If l r 0 is given, why do we need to remove the first index greater than $r$ ? Shouldn't it be "remove the last index with the value at most $r$"? Cmiiw
•  » » 2 years ago, # ^ |   0 Oh nevermind. I wrongly understand the statement. It means "keep deleting the lower_bound until we find a value greater than $r$"
 » 2 years ago, # |   -24 Please help me in understanding the editorial of problem div2E (1641C - Анонимность наше все). Or share some alternate approach that is easy to understand.
•  » » 21 month(s) ago, # ^ |   0 Have you understood it? I am also confused.
 » 2 years ago, # |   0 I am totally confused about why my submission at problem C got WA at 42nd testcase in case 2I get '18' while the Jury gets '17'
•  » » 2 years ago, # ^ |   0 I don't clear that array completely
 » 2 years ago, # | ← Rev. 2 →   -8 May I know how so many people got TLE'd on 1642C - Великая последовательность, especially on the systests? I am suspecting that people went for adding numbers instead of erasing and that added to the runtime, but I am not quite sure.
•  » » 2 years ago, # ^ |   -11 (sure, I do understand how the problem had weak pretests, but may I at least know why these solutions got TLE'd on the main tests only? I don't think there would have been an edge case that would make the runtime so long either.)
•  » » » 2 years ago, # ^ | ← Rev. 2 →   0 it's because many people used unordered map to solve the problem. umap in c++ generally has constant insertion and lookup times, but there are values that can cause collisions in the hash, thus causing a worst case time complexity of O(n^2). you can read more about it here https://mirror.codeforces.com/blog/entry/62393
•  » » » » 2 years ago, # ^ |   0 will it be okay if i use map instead everywhere? can map give tle in cases where umap would not? if yes then how to tell?
•  » » » » » 2 years ago, # ^ | ← Rev. 2 →   0 yeah, it can, just because there are some scenarios where a map's log(n) lookup and insertion are slower than umap. I havent come across many problems where its a big deal though, you can probably just go off of intuition if you're not sure if your code will tle with map. (but typically, it's a better idea to use map than umap if it doesn't make a big difference)
 » 2 years ago, # |   +5 Cnacu6o for weak pretests of D( ya nje ljubju FST
 » 2 years ago, # |   0 Figured out the solution of 1642D - Расшифровка повторяшек during the contest but can't put it into code in time... so sad
 » 2 years ago, # |   0 Why I got MLE at Problem C at the 10th testcase? I used map to count the numbers, the total memory complexity may be O(n). 147431101
•  » » 2 years ago, # ^ |   0 Problem solved.
 » 2 years ago, # | ← Rev. 2 →   0 I thought of the same thing in prob D but failed to come to the point that by doing this we can sort it
 » 2 years ago, # |   +21 I got a different solution for Div. 1C just after contest ended. (https://mirror.codeforces.com/contest/1641/submission/147504174) Let's call each query's index as its timestamp. Basic idea is to find out the earliest timestamp the status of each index becomes determined. First, process all t = 0 && x = 0 queries. Record the first time each index becomes 0. Maintain a segment tree with point update min. Second, process all t = 0 && x = 1 queries. We know that for a query with l and r, if and only if there are r-l 0 indices inside it, the left one is determined to be 1. So query the previous segment tree for range sum could find out this. Notice also need to find out the largest timestamp of all 0 indices. The maximum of 0 indices and this 1 query is the timestamp of this 1 index. Notice that, there might be more than 1 queries making this index determined, and we need to pick the minimum one. Last, process all t = 1 queries. if the index has determined status, and the timestamp is earlier than the query, then result is YES or NO, otherwise N/A.
•  » » 2 years ago, # ^ | ← Rev. 2 →   0 I also had a similar solution — (https://mirror.codeforces.com/contest/1641/submission/147472944)Sadly couldn't debug within time:(
 » 2 years ago, # | ← Rev. 2 →   0 For Problem A, if I try using long double, I get a wrong answer, but if I use long long, then I get correct answer. Why does this happen? My logic is the same as the editorial.
 » 2 years ago, # |   +67 d1D can be solved in $O(\frac{n^2m}{\omega})$ using bitset.d1E can be solved in $O(n^2)$ using Ofast.Great contest!
 » 2 years ago, # |   +10 I am a idiot. I forgot to printf the answer in problem E so that I can't solve it.
 » 2 years ago, # | ← Rev. 2 →   0 Div2C seems to have weak tests guys. The following submission should TLE. https://mirror.codeforces.com/contest/1642/submission/147435078But its AC. I used a multiset with count, which is slow (linear time complexity). Sample input used 1 200000 1 (100000 times) 2 (100000 times) 
•  » » 2 years ago, # ^ |   0 Hi,I tried to hack your solution with the case that you have described (Hack #788603). Your solution runs in time.Generator Used: Spoiler#include using namespace std; int main() { ios::sync_with_stdio(false); cin.tie(nullptr); cout << 1 << endl; const int M = 200000; cout << M << ' ' << 2 << endl; for (int i = 0; i < M / 2; i++) cout << 1 << ' '; for (int i = 0; i < M / 2 - 1; i++) cout << 2 << ' '; cout << 2 << endl; return 0; } 
•  » » » 2 years ago, # ^ |   0
•  » » 2 years ago, # ^ |   0 I tried to experiment your solution in my local machine. It only TLE (runs > 1s) if I turn optimization (-O2, default in Codeforces) off. The compiler is smart enough to optimize out such inefficiency. :)
 » 2 years ago, # | ← Rev. 2 →   0 Can someone please help me understand why I am getting TLE for these submissions in Div2-B. I have used unordered_map and unordered_set. (147532491 and 147456652).
•  » » 2 years ago, # ^ | ← Rev. 6 →   +7 Never use unordered map in codeforces (or in any other contest). Always use ordered_map (map in c++).The reason is the worst case time complexity for both unordered sets and maps is O(n). And you can bet the testers would have created a test case just to exploit exactly that.There is no downside to using ordered maps or sets, since the worst case time complexity is O(logn) which is almost never a problem. (log2(10^5) ~ 17, so your search is complete in nearly 17 steps. Running 17 steps is a joke for cf servers and other online judges.)
•  » » » 2 years ago, # ^ |   0 Why? Is there some issue with the compiler? Also what is wrong for unordered_set?
•  » » » » 2 years ago, # ^ |   0 see the edit above
•  » » » 2 years ago, # ^ |   +11 What's wrong guys? Why the downvotes?
 » 2 years ago, # |   0 Can anyone explain me why unordered_set or unordered_map gets TL for the problem B but set/map doesn't get TL? submission 1 : 147535731 submission 2 : 147535959
•  » » 2 years ago, # ^ |   0 The answer is just above. Just scroll up. https://mirror.codeforces.com/blog/entry/100249?#comment-889850
•  » » » 2 years ago, # ^ |   0 Thanks :) I thought it's almost impossible to get O(N) using unordered_set/map. I don't know properly how it works behind the scenes
•  » » » » 2 years ago, # ^ | ← Rev. 3 →   0 It is almost impossible to get O(N) using unordered_set/map in the real world. You are not wrong. It's just that cp is not the real world and the testers happen to know about the weaknesses of each data structure much more than we'd like :)There are testers who specialise in creating adversarial test data just to break the hashing strategies of unordered map/set and force it run in O(n). Thankfully such kind of test data is ineffective against an ordered map.
•  » » » » » 2 years ago, # ^ | ← Rev. 3 →   +10 Guys if you are going to downvote my comment, at least explain why. I tried to explain something to the best of my ability, if I am wrong let us know why.
 » 2 years ago, # | ← Rev. 2 →   0 ilyakrasnovv correct me if I am wrong, but this line in explanation of Div1C -we can find the first elements to the left (l) and to the right (r) of i, which might be healthy using our set. The i-th person is ill when the minimal element on segment [l+1;i] is
 » 2 years ago, # |   0 I'm a freshman ,i used multiset in 1642C,but i got TLE , can it be solved used by multiset only? can somebody help me , thanks a lot! It's my submission 147572794
•  » » 2 years ago, # ^ | ← Rev. 2 →   0 There're some repetitive computation in your code. for example, both s.count(a[i]) and s.find(a[i]) search a[i] in s, same for s.count(x * a[i]) and s.find(x * a[i]).Here's the AC code I modified based on your code: https://mirror.codeforces.com/contest/1642/submission/147613781Actually, since multiset is an ordered structure, we don't need to sort vector a. we can just pick every element in the set directly until the set is empty.here's my code based on this idea: https://mirror.codeforces.com/contest/1642/submission/147614303 :)
•  » » » 2 years ago, # ^ |   +10 谢谢！ 学到了很多！
 » 2 years ago, # |   +40 I think i've found a nice randomized solution to D1D.First compress all the values. Now generate a random permutation and replace the corresponding values with it. Now lets check all the pairs such that the minimum value in one array is bigger than the maximum value in the other. This works with probability of $1 / C_{2m}^m$, which is $1 / 252$ for $m = 5$. So now all we have left to do is process this in linear time ($O(nm)$) and fit into TL:) I would consider the total time complexity to be $O(n\cdot m\cdot C_{2m}^m\cdot log(error)) \approx O(n\cdot m\cdot 2^{2m}/\sqrt{2m} \cdot log(error))$.Solution: 147616242
•  » » 2 years ago, # ^ |   +36 I have similar randomized solution.Let $W=13$, then we can produce randomized function $\varphi$ that maps bits from 0 to 12 to each natural number. Next, we can match each array with a bitmask equal $\Phi = \varphi(a_1) | \ldots | \varphi(a_m)$. Thus, in one iteration, it is enough to consider pairs of arrays whose masks do not intersect in $O(N + W \cdot 2^W)$ time. We have to do about 100 iterations of this algorithm for the correct answer.
•  » » 2 years ago, # ^ | ← Rev. 2 →   0 I tried same approach but always got WA. Maybe my implementation has big constant factors. If we do $1000$ iteration in TL, we will have about $0.98$ probability to get a correct answer. But it's not enough to get AC. With $0.98$ probability per test, we only have $0.13$ probability to get a AC(pass $100$ tests), which is still need some good luck.
•  » » » 2 years ago, # ^ |   0 That's actually trueSolution of teraqqq is way better in this aspect
 » 2 years ago, # |   +24 Correct me if I am wrong, but this line in explanation of Div2E/Div1C: SpoilerIf $i$-th person is ill, the following query exists: $0 \ l \ r \ 1$, such that $l \leq i \leq r$, and every person $i$ such that $l \leq i \leq r$ are not ill. If there is such person $j$ that they are not ill, and $j \neq i, l \leq j \leq r$. In this case, it is impossible to determine if i-th person is ill or not.should be: SpoilerIf $i$-th person is ill, the following query exists: $0 \ l \ r \ 1$, such that $l \leq i \leq r$, and every person $j$ such that $j \neq i, l \leq j \leq r$ are not ill.
•  » » 2 years ago, # ^ |   0 I didn't understand the last (segment tree part) part of the editorial. How can we find a person is definitely ill using seg tree? Can you help me? Thanks in advance.
•  » » » 2 years ago, # ^ | ← Rev. 3 →   +5 Otherwise, we can use a segment tree to store such index $j$ that there is a query $0 \ i \ j \ 1$ and store it in the i-th slot of our segment tree.When we understand that the i-th person might be ill, we can find the first elements to the left ($l$) and to the right ($r$) of i, which might be ill using our set. The i-th person is ill when the minimal element on segment $[l+1;i]$ is $•  » » » » 2 years ago, # ^ | +5 Got it. Thanks a lot. •  » » » » 6 months ago, # ^ | 0 can it be solved if question was modified to allow$l≤b
 » 2 years ago, # |   +5 “when the minimal element on segment [l+1;i] is
 » 2 years ago, # | ← Rev. 2 →   0 In Div 2, C 1642C - Великая последовательность why does the editorial solution have an if statement for when x is 1, if it is specified in the problem statement that x is at least 2?
 » 2 years ago, # | ← Rev. 2 →   0 in div2 problem C, I got it wrong at first then it felt right to sort the array and got it AC. the thing I myself am not fully convinced why sorting was necessary. can someone explain or some proof of thought(not any formal proof, just the idea of correct thinking).
•  » » 2 years ago, # ^ | ← Rev. 2 →   0 Try the following case with x = 2: 2 4 1 8
 » 2 years ago, # |   0 Problem 1642E — Anonymity Is Important. I didn't understand the last (segment tree part) part of the editorial. How can we find a person is definitely ill using seg tree? Can someone explain? Thanks in advance.
 » 2 years ago, # | ← Rev. 2 →   0 Can somebody explain why in div.2 E the observation that "$i$ is ill when minimal value in our segment tree on interval $[l + 1; i]$ is smaller than $r$" is working?
•  » » 2 years ago, # ^ |   0 It's easy to understand that : if there exists a segment [x,y](when x in [l+1,i], y in [i,r-1]) , the i-th person is ill. Then let's consider the segment [x,y] whose x is in [l+1,i]. There is no ill person in [l+1,i), so y>=i. So "if there exists a segment [x,y](when x in [l+1,i], y in [i,r-1])" is equal to "if there exists a segment [x,y](when x in [l+1,i], y
•  » » » 2 years ago, # ^ |   0 Thank you SO MUCH for your help, after your explanation I started experimenting and finally understood what they meant in the editorial. Sorry for my late response.
•  » » » » 2 years ago, # ^ |   0 you are welcome！
 » 2 years ago, # |   0 div2 F's solution is so cool!
 » 16 months ago, # |   0 I'm trying to solve div-1 C with O(n*(log n)^2), is there any way to optimise it a bit so that it wont tle?https://mirror.codeforces.com/contest/1641/submission/192119258