Hello Codeforces!
mesanu, flamestorm and I are very excited to invite you to Codeforces Round 859 (Div. 4)! It starts on Mar/19/2023 17:55 (Moscow time).
The format of the event will be identical to Div. 3 rounds:
- 5-8 tasks;
- ICPC rules with a penalty of 10 minutes for an incorrect submission;
- 12-hour phase of open hacks after the end of the round (hacks do not give additional points)
- after the end of the open hacking phase, all solutions will be tested on the updated set of tests, and the ratings recalculated
- by default, only "trusted" participants are shown in the results table (but the rating will be recalculated for all with initial ratings less than 1400 or you are an unrated participant/newcomer).
We urge participants whose rating is 1400+ not to register new accounts for the purpose of narcissism but to take part unofficially. Please do not spoil the contest for the official participants.
Only trusted participants of the fourth division will be included in the official standings table. This is a forced measure for combating unsporting behavior. To qualify as a trusted participant of the fourth division, you must:
- take part in at least five rated rounds (and solve at least one problem in each of them),
- do not have a point of 1400 or higher in the rating.
Regardless of whether you are a trusted participant of the fourth division or not, if your rating is less than 1400 (or you are a newcomer/unrated), then the round will be rated for you.
Many thanks to all testers: jampm, Max_Calincu, KrowSavcik, TimaDegt, nyaruhodo, tibinyte, badlad, Phantom_Performer, AlperenT, Bakry, keta_tsimakuridze, Gheal, RedstoneGamer22, Dominater069!
And thanks to Alexdat2000 for translating the statements!
We suggest reading all of the problems and hope you will find them interesting!
Good Luck to everyone!
UPD: Editorial
What will I gain from hacking if there are no points?
you have chance to higher your place
I do it because I like the +100 next to my name. By the way, if you want to know a safe solution to G2, check out my solution video.
thank you i'll watch the video , because i don't know what's the solution of E and F.
I watched your video, it was really simple explaination, but for some reason I was not able to AC G2. https://mirror.codeforces.com/contest/1807/submission/198285267
What is wrong? Could anyone take a look?
For WA in test 14, I guess it's the integer overflow problem
Getting TLE in test 25
Submission: https://mirror.codeforces.com/contest/1807/submission/198325433
don't use Arrays.sort(), in worst case it gives O(n^2). I did the same mistake TT. Make your own sort() function or try to use Collections.sort().
super villain
Increase your hacking ability and it's fun (:
it's not fun for the person hacked :(
clout.
Marinush
Sorry, what is hacking actually?
hacking is to give a testcase that makes a solution to a participant fail.
example : someone solution get passed during contest but after the contest over you can try custom testcase which won't pass the solution and it will get w/a and you'll get some point for every successful hacking.
That is evil!
But some guys can gain more point.and i'm often hacked by others,i can understand you
You can ruin someone's evening!
U R right!
As a tester, I suggest you participate! Problems are nice and educational!
I love div4 contest because i get positive delta.thanks for doing div4 contest!!!
I wish to see nice problems like other div4s
As a participant, I want to thank you for the contest!
What I was waiting for the most!
loh
ooh div.4 . Means SlavicG in action :)
OMG SlavicG Round!
As GPT-4, tomorrow will be my 5th contest participation. I hope I become green :)
Added you as a friend, already. Please, note, that while all the rest were laughing at you, I was always on your side! Hope you'll kill me last.
Going to be my first unofficial round.
flamestorm orz
people on their way to post this meme hoping they will get some upvotes
Div 4
So, easily will do 80% of the problems.
As a tester, Give me contribution!
Hope to become expert in this contest.
funny !
Hope to see the colour change in this round !!
( Δ > +6 )
>=
most probably... atb!
hello. i am man; i am from to mars and run away at abi abi is brother but not brother; abis isnt cool; int monkey; monkey = monkey + wepon; cout << monkey kill nuraly SH;
Why are you so talented????????????????
I'm not talented, it's just practice, practice makes you talented
stop yapping
Hey, are you a contest? 'Cause you're looking CUTE!
I would smash this contest
Delayed round?
Starting time 10 mins. extended...
Div 4 contest, more like load testing for Codeforces Servers
2 tests failed, +20 minute penalty
the contest keeps getting delayed by 10 minutes for me? just me?
Not just you,.
LateForces
More than 30k participants and now Codeforces is trafficked
Last Div4 was 38k ....
Today i'm grey again..
DelayForces
Feeling like participating in onsite contest where after every refresh 10 minute increase.
Worst Server I Have Ever Seen!..
Div.4 is good for me~
May be, the queue more long than my imagination :3
)
If there was going to be interactive problem.It should be mentioned in blog beforehead.
Interactive problem isn't really that different from normal problem.
that long queue really sucks
Didn't know for like 10 minutes my solution was too slow...
QueueForces:(
In queue forces :(
There must be a prior information about Interactive problem in contest. i have seen several div 2 contest where author has given prior info about interactive problem.
Where in codeforces rulebook does it say that? Why does everyone hate interactive problems?
I like the contest, but the website is not co-operating :(
I think it was supposed to be mentioned before the start of the contest about the interactive problem.
I'm praying to the universe this contest does not go unrated :)
Seems like Mike is manually judging solutions
Lmao
I apologize for the long queue at the round. Today's round was a record in terms of the constant huge flow of submissions. Unfortunately, we ran into the speed of compilation. I already have plans to move the compilation somewhere to the cloud in such extreme cases. I plan to implement this for the next div4.
bits/stdc++.h showing it's true power
please make this round unrated it wasn't a normal Codeforces round queue was way too long T_T
It was fair because everyone was at a disadvantage. Everyone's queue was 10 minutes long, I don't see how it affects one's ability to problem solve.
Interactive problem + long queue = nice COMBO!
The long queue has taken a toll on those of us who rely on proof by AC.
solving 1 (last problem took me more time than solving all the rest of the problems combined...
Due to WA and LONG WAITING QUEUE.
Lagforces...
Are pretests on hard version of G weak? I realized my code was doing the opposite of what I wanted it to do after submitting and it still got AC... I have no idea how my solution works.
Bruh why did you let bitset pass G2?
UPD: I got hacked lol
div4Forces
hacked uwu
I solved 1807G2 - Subsequence Addition (Hard Version) with C++
bitset
too. At current time I have800 ms
.My submission.
I see, that iiand tries to hack it right now. Hope, that he will share test case when succeeded, it is quite interesting.
Perfect solution:)
I thought all this time that this is bitset-related problem and solution with bitset is intended solution. Was really surprised when discovered alternative greedy solution in comments under this blog...
Very hard to track if the submitted answer was right or not !!
A lot of time wasted there. Apart from that really very goooood contest.
Любительский разбор задач
Проверьте, что
a+b==c
. Самый короткий способ:cout << "-+"[a+b==c] << '\n';
Ссылка на AC сабмит
Разместите чётные конфеты все до нечётных. Выяснится, что самое последнее сравнение будет "сумма всех чётных" минус "сумма всех нечётных" и можно проверить именно его результат. Остальные более благоприятные для нас и их можно не проверять.
Ссылка на AC сабмит
У нас есть всего две чередующиеся бинарные строки длины n: одна начинается с $$$0$$$, вторая начинается с $$$1$$$, то есть $$$01010101...$$$ или $$$101010101...$$$ Проверим каждую.
По условию задачи, мы каждой латинской букве сопоставляем $$$0$$$ ИЛИ $$$1$$$. Давайте проверим каждую букву. Если есть такое, что одна и та же буква должна перейти в $$$0$$$ в какой-то позиции и в $$$1$$$ в какой-то другой позиции, то ответ
No
. Если ни одной такой буквы нет, то ответYes
.Проверку можно провести следующим образом. Заведём
map<char,int>
— какую цифру мы сопоставим каждой букве. При проверке позицииi
проверяем, есть ли букваs[i]
в мапе как ключ (сопоставили ей что-то или нет). Можно черезmap.find
илиmap.count
илиmap.contains
. Если нет, то сопоставляем, иначе она есть и это будет означать, что она уже встретилась раньше. Проверим, что до этого мы сопоставили ровно то, что хотим сопоставить сейчас. То есть, если мы хотим $$$1$$$, то в мапе лежит именно $$$1$$$, а не ноль.Ссылка на AC сабмит
Построим префикс суммы поверх массива. Теперь умеем считать сумму на любом отрезке за $$$O(1)$$$. Обрабатывать запрос за $$$O(1)$$$ можно следующим образом: возьмём предподсчитанную сумму всех чисел $$$S$$$, вычтем из него
(pref[r] - pref[l-1])
и прибавим(r-l+1LL)*k
. Именно так изменится сумма после замены подотрезка. Если получилось нечётное число, то ответYes
, иначеNo
.Ссылка на AC сабмит
Изначально в любой кучке может быть загаданный камень, то есть, кандидаты
{1,2,3,...,n}
. Давайте делить множество кандидатов пополам (на две группы примерно равной длины). Например, все кандидаты на чётной позиции массива относим в первую группу, а нечётной — во вторую группу.Тогда мы можем спросить сумму в первой группы: какая она? Нам сообщат число. Давайте проверим, что оно реально сумма в этой группе (мы же знаем индексы). Если всё совпало (оказалось, что в первой группе нет нашего камня), то ответ режит во второй группе, иначе ответ лежит в первой группе.
Таким образом, на каждой итерации мы уменьшаем множество подходящих ответов вдвое, пока оно не станет содержать ровно один элемент. Его мы и выводим в качестве ответа.
Ссылка на AC сабмит
Суммарное количество клеток поля, по условию,
n * m <= 5 * 10^4
. Это $$$50$$$ тысяч клеток. В каждой клетке мы можем находиться в четырёх состояниях (в зависимости от текущего направления). Получается, у нас $$$50000 \times 4$$$ состояний в худшем случае и мы между ними переходим.Можем промоделировать явно, начиная со стартового состояния, пока не придём в искомую точку.
На каждой итерации, если мы уже в искомой точке, то сразу сообщаем ответ и заканчиваем моделирование, иначе надо сделать шаг.
Посещённые состояния будем хранить в виде четвёрки чисел (строка, столбец, смещение вверх, смещение вправо). Можем заносить их в
std::set
. Если в некоторое состояние мы пришли повторно, то завершаем моделирование процесса.Отскок проверяем так: если вышли за пределы карты (по строкам и столбцам отдельно), то возвращаемся и смещение меняем на противоположное (по строкам и столбцам отдельно) и делаем шаг в изменённом направлении.
Получается $$$O(n \cdot m \cdot \log{n\cdot m})$$$.
Ссылка на AC сабмит
Будем набирать элементы в порядке их неубывания (сортим вектор $$$c$$$). У нас уже есть
{1}
, поэтому, еслиc.front() > 1
, то ответ сразу No, иначе набираем элементы начиная со второго.Элементы
c[i]
до $$$200000$$$. То есть, наша сумма подпоследовательности до $$$200000$$$. Давайте хранить все возможные суммы — их всего-то $$$200000$$$. Где хранить?Давайте заведём битсет на $$$200200$$$ элементов, где
bitset[s] = 1
, если подпоследовательность с данной суммой s уже есть, иначе $$$0$$$. Изначально толькоbitset[1] = 1
.Итак, пусть мы хотим набрать
c[i]
для каждого $$$i$$$ от $$$2$$$ до $$$n$$$. Сперва проверим, можем мы его набрать или нет. То есть,bitset[c[i]]
равен $$$1$$$ или нет. Если не можем, то ответ No, иначе можем набрать, набираем (добавляя в наш массив $$$a$$$). У нас появился новый элемент в векторе $$$a$$$, равныйc[i]
. Давайте обновим все возможные подпоследовательности. У нас уже они были без учёта этого элемента. Теперь в каждую из них мы можем добавитьc[i]
или оставить без изменений. То есть,bitset |= (bitset << c[i])
— обновление множества подпоследовательностей за $$$O\left(\frac{n}{32}\right)$$$.Итого, получили решение за $$$O\left(\frac{n^2}{32}\right)$$$. Оно работает за
800 мс
(с запасом в $$$2.5$$$ раза).Ссылка на AC сабмит
Блин, клевый способ для G. Я решил жадно и сортировкой, но это прям рабочий такой способ! Спаисбо.
Я уверен, что в разборе будет какой-то жадный способ, однако считаю, что для див. 3 и 4 разборы должны содержать именно вот такие решения как ваш разбор для G, потому что какие-то "элегантные" решения -- это, на мой скромный взгляд, самое трудное и, лучше приводить более рабочее решение, к которому можно трудно, но прийти. Нежели если я там на уровне интуиции и поверхностных доказательств буду сидеть 2 часа и рожать решение.
Спасибо за разбор! Непонятно, как пользоваться этим bitset? Что делает операция bitset |= (bitset << c[i])?
bitset<N>
— набор из $$$N$$$ бит, каждый из которых может быть $$$0$$$ или $$$1$$$.bitset << k
— побитовый сдвиг вправо на величину $$$k$$$. В терминах суммы, это прибавление $$$k$$$ во всем существующим суммам (индексам битсета).left | right
— оператор побитового ИЛИ. Выставляет единицу по индексу[i]
, если хотя бы один из операндовleft[i]
илиright[i]
равен $$$1$$$, иначе $$$0$$$.Для
bitset
реализованы все побитовые операции. Рекомендую ознакомиться с его возможностями в туториалах в интернете и на cppreference.а почему О(n^2/32) заходит по времени? кажется что n^2/32 > 1e9. почему такая оценка не верна, из-за того, что побитовые операции быстро работают?
Константа низкая. У серверов Codeforces тактовая частота 3.6 GHz вроде. 3.6e9 тактов в секунду. Решение потратило всего 2.8е9 тактов. Предполагаю, что предсказуемое последовательное чтение из памяти, сдвиг, побитовое ИЛИ + предсказуемая последовательная запись в память это пара тактов процессора в среднем, тем более с включенным набором Ofast+avx, который подрубит 256-битные YMM или 128-битные XMM регистры в зависимости от предпочтений компилятора
спасибо!
How did I get G1 accepted and G2 WA with the same solution provided
just use long long
My advice is if you use c++, add
#define int long long int
that is not a good advice, you can TLE if you're not careful, I advice you to manage your limits properly instead.
I suppose, but that has never happened to me before.
Yes, many of the setters are pretty generous with their time limit, and you should easily pass even with long long, this is not always the case though, better to be careful
You can get MLE or TLE with that in some problems.
Amazing Problems <3
I loved E because It was first time I solved an interactive problem.
Only if Queues werent that long I could have figured out my Idleness Limit Exceeeded 10 minutes before :(.
Solved everything except F
nice smash hulk
Haha Thanks
Contest over,still my solution is in queue :))
I am using prefix sum in G,why it is giving WA My solution
if(n == 1 && v[0] > 1){ pn; return; }
you dont need n==1 here because you cannot change all 1s in the initial array.
How to solve G2?
if the kth element of the sorted array is less than or equal to the sum of all 0...k-1 elements then ok else no. iterate for all from 1 to n-1 (0 based) and sorted
It is always correct to add numbers in increasing order. So you can sort
a
and there holds invariant: ifa[i] > a[0] + a[1] + ... + a[i - 1]
then there is no way to adda[i]
. And ifa[i] <= a[0] + ... + a[i - 1]
then you can adda[i]
and you can add fora[i + 1]
any number from1 to a[0] + a[1] + ... + a[i]
.So just sort and check that for each
i a[i] <= a[0] + ... + a[i - 1]
" And if a[i] <= a[0] + ... + a[i — 1] then you can add a[i] "
How?
Let's assume that you processed indexes
0..i
, and you established that you can always get any sum from1
toS = sum(a[0..i])
. Then there are two cases:a[i + 1] > S
then you cannot adda[i + 1]
and answer is NO.a[i + 1] <= S
then you can adda[i + 1]
and you can get any sum from1
tosum(a[0..i+1])
(because you can take any sum from1
toS
in indexes from0
toi
, and you can adda[i + 1]
to that).Ok, so you are using induction.
G1-G2 were easy when u observe that as we have the minimum unit as 1 so we can form any possible sum. So say we sort the array now following conditions should hold.
" when u observe that as we have the minimum unit as 1 so we can form any possible sum "
Could you please explain how is it possible to form any sum?
Although minimum unit is 1, but we need to make sure that intermediate values should also be there in a.
As we are given, $$${a = [1]}$$$.
So by this, you can observe that in any way we can form the next number $$${1, 2, 3, 4}$$$ by choosing any above combinations.
So via this, you can prove any next number is possible if the prefix sum is greater or equal to that number.
Thank you
You can prove it by induction. Lets say I can make any sum from $$$1$$$ to $$$k$$$ using the current numbers from $$$a_0,a_1,\cdots,a_i$$$. Then consider $$$a_{i+1}\le k$$$. Since I can make any sum already from $$$1$$$ to $$$k$$$, I can make $$$a_{i+1}$$$ and add it to my list of numbers, and now if I add $$$a_{i+1}$$$ to those previous sums, I can now make any sum from $$$1$$$ to $$$k + a_{i+1}$$$. However, if $$$a_{i+1}\gt k$$$, then notice that I can never make $$$a_{i+1}$$$, because the current numbers cannot make any sum greater than $$$k$$$.
So if at any point $$$a_{i}$$$ is greater than the sum of the numbers before it in sorted order, the answer is NO.
Simple solution: Sort array, go over each number and make sure its not greater than the running sum, unless the number is 1.
beautiful problems, even if i couldnt solve all i was manage to at least have a go at it. Perfect div.4 round
Yes. Perfect Div4 round indeed. Was able to solve only A-E during contest, but first time I upsolved all questions of any contest.
Solution for F? I got TLE :D
F was terrible enough!
It was completely based on implementations. I myself wrote $$$696969$$$ lines of code.
My submission
Most of them are your include :)
PrefixsumForces
YESNOForces
https://mirror.codeforces.com/contest/1807/submission/198254663 why T.L.E
you can refer my solution :https://mirror.codeforces.com/contest/1807/submission/198219539
Good contest! Wasn't too fond of F though, so didn't do.
Waited 15min to get wa on test 1 (E problem).
In fact, testing was slow today. But I'm not sure you're right about 15 minutes. What is the id of your submission?
E had the longest queue out of all my submissions for some reason. Waited so long to get WA on both sucked when I could have used the time to solve F
It looks like 10, but not 15 if you want to round it. It took a long time to test (sorry about it), and I'll work on it. But you exaggerated almost one and a half times.
For G2 and G1, after sorting the array I knew that that one of the conditions was that every element should be at max the sum of all the elements that occured before it. Little did I know that it was the only condition required. Could someone explain me why we can always make every number from [1 — sum of all elements occured] from the given elements ?
We can prove by induction...
Obviously it is possible to reach all numbers that are less than or equal to 1 by using 1. Then consider a time when we add a number j and the sum of existing elements is i. Assuming that all numbers up to i can be reached. Then all numbers from i to i + j can be reached by adding j to a number from 1 to i.
F will be an interesting (and definitely too hard for div4) problem if the constraints are n,m<=1e9 and no guarantee for their sum.
That would've been one of the best geometry problems.
Did something similar 198269402
Why the following code for D is getting TLE? It has time complexity of O(n+q)
Use C++ or fast IO
Even after many attempts , i am unable to remove "Idleness Limit Exceeded" to the problem E : 198257666.
Your help is appreciated.
endl in c++ flushes the i/o stream by default so no need to use cout<<flush after using endl Though I didn't check your code thoroughly (some other bug maybe there)
i did the same but still it gives the same error.submission
Remove the
n == 2
case. query output format is wrong and this case is not needed at all.Can someone please explain how to do problem D within time limits? And what optimization was needed for G2, i got G1 correct, sorted the array and checked if any element was larger than the sum of those before it, but TLE on G2 at test 19(coded everything in C)
using prefix sum. .. pre compute the prefix and suffix sum of the array. so for range[l,r] we need the sum of elements excluding this range which can be calculated using the pre computed sum. prefix[l-1]+suffix[r+1]and for the range sum can be calculated as k*(r-l+1)..
I found the sum of all elements once and then just subtracted the elements in that range for each query and added k*(r-l+1), won't that be faster than calculating prefix sum for each case https://mirror.codeforces.com/contest/1807/submission/198251225
no, clearly not, prefix sum calculation takes o(n), and query takes o(1), the worst time complexity if you calculate for each is o(nq), while if you precalc, its o(n+q)
What is wrong with my solution of D.?? Please look into it. https://mirror.codeforces.com/contest/1807/submission/198208523
int overflow where you used --->>> (r-l+1)*k <<<----
so what should i do in its place take individual remainders and then add
Use "long long"
please can someone tell me why is my solution to problem E giving WA on testcase 1? https://mirror.codeforces.com/contest/1807/submission/198264987
you printed the "value" at the indices in the queries, we had to print "indices" in the queries
change v[i] to i
thanks bro, im so stupid xdxd
I see someone using correct way to do G2 but it hacked...
a[0] must = 1 for all 1 <= i < n, if a[i] > sum(a[0]~a[i-1]) then answer is NO. else is YES
is it wrong?
I think the general idea is right (it's what I did anyway). Which submission are you referring to?
Problem E Flushing Problem what's the wrong with using '\n' always (Idleness limit exceeded on test 1), using endl (Accepted)
I submited with '\n' and it passed.
ios::sync_with_stdio(false); cin.tie(NULL); cout.tie(NULL);
cout<<"! "<<s<<'\n';
fflush(stdout);
No! First one unties cin/cout operations from stdin/stdout operations, so it is implementation-defined (or even undefined, but no sure) behaviour to mix up them both after that.
The problem F was not hard to understand but hard to implement. My guess is that there is probably a great implementation that does not require bunch of if and else blocks.
Can someone share a nice and compact implementation or the idea behind it?
I think my implementation is not bad though
https://mirror.codeforces.com/contest/1807/submission/198179860
F with simulation
How on earth Yodasen made possible to write solution and submit within the time interval of 5 second (solution A(3 min 53s) & D(3 min 48s)) and 21 second (solution C(7 min 30s) & G1(7 min 9s)) second!
Is this real??
Also, look at his submissions. The templates are different, obvious 2/3 people on a single account.
For problem F, while contest I didn't read constraints properly, I designed more generalised solution. my solution will work for N <= 10^5 ( or even N <= 10^6 ) .
[submission:https://mirror.codeforces.com/contest/1807/submission/198280396] .
I don't think you solution is a more generalized version. If the pigeonhole thingy from your solution actually holds, then it holds in everyone's solution that memoized too.
most of the solutions are using dx[4] = {1,1,-1,-1}, dy[4] = {1,-1,1,-1} , and ball is moving one by one step( from cell (i,j) to (i+1,j+1) or (i-1,j-1) .. etc ).
In my solution, ball jumping from one wall to another wall in O(1) time.
Pigeonhole principle will hold only for boundary cells. Not for inner cells.
I haven't seen any solution moving cell by cell, they will probably TLE if they do that. Every AC solution I've seen moves in O(1) time between the walls.
mine moved cell by cell and it work fine, no TLE
You should really check before making incorrect claims. btw, these were the first 8 submissions I checked.
oh no, another ScarletS notification. leave me the fuck alone.
Did you even read his initial comment? How is that a generalized solution?
I said I haven't seen any AC solution using that, and they will PROBABLY TLE, so what is the wrong claim there???
oh no, another instance of Trippie saying dumb shit, misleading people and expecting to not be corrected!
And I said nearly everybody submitted such a version, and pulled up the first 8 submissions I checked, implying that you probably didn't check.
Keep spreading nonsense from an alt though, I guess noone would take you seriously on your main anyways.
Where did I mislead anyone? Lol
Go through the first page of the submissions, and more than half of them don't go through cell by cell.
The Div1 guy is always right anyway, enjoy your internet points.
First submission on the page: 198531289.
In this case, sure. Keep digging yourself a hole though.
Man, how stupid are you? I said more than half of them did I say the very first one?
Pretty much every single submission on that page is cell by cell. Stay grey though.
Lol, how'd you figure I'm grey? Oh, no! Anyway, this conversation isn't going anywhere and has diverted from my initial argument. Ths solution wasn't a generalization, and I thought the solutions moving cell by cell will TLE that's why I used the word PROBABLY (you can use a dictionary). Have a bad day!
The right move would be for you to admit you were wrong and move on. Keep digging your hole though!
Hello, why is my code for problem D giving WA on test case 4? The logic should be correct and I cannot think of any bugs. Is there something I have failed to account for? There shouldn't be any issues with int overflow I believe.
Code
Thank you all very much in advance.
Your code outputs "NO" instead of "YES"
The error is that l, r, and k are ints, so temp += (r-l+1)*k first evaluates (r-l+1)*k as an int, which overflows.
I think (r-l+1)*k can overflow
how is e related to dp???
if you sort array, then:
dp[i][s] = if you can make sum S using the first i elements
base case:
dp[1][1] = 1
(and check ifa[1] == 1
)transition is:
dp[i+1][s] = dp[i][s] | dp[i][s-a[i]]
Observation is that if some sum
t
can be created using firstp
elements, then it's also possible to create any sum<= t
. So we can remove one dimension from the dp and store only the maxs
for whichdp[i][s] = true
.This is actually how my reasoning went during the contest.
Can you prove this observation?
proof by induction:
Define
s[i]
max sum that can be created from firsti
elements.It's true for
i == 1
, becausea[1] == 1
and it's possible to create sums[1] = 1
onlyAssume it's true for i, now prove that it's also true for i+1. If
a[i+1]
is greater thans[i]
, then the answer isNO
, so assumea[i+1] <= s[i]
. There are 2 cases. To create sumx <= s[i]
we can do so without including current element, because it's already possible to do so using previous elements. To creates[i] < x <= s[i] + a[i+1]
we can use current element, and create sumx - a[i+1]
from previous elements. Creating sumx > s[i] + a[i+1]
is impossible, becausex - a[i+1] > s[i]
and (by definition)s[i]
represents maximum sum that can be created using firsti
elements. Sos[i+1] = s[i] + a[i+1]
Forgot to mention why sorting is necessary:
A number can only be constructed using numbers less or equal to it, because no negative numbers are allowed
How does one become a tester?
Can anyone explain why am I getting TLE in the given code below.
https://mirror.codeforces.com/contest/1807/submission/198262166
I am not sure but I think because the worst thing for your code to not find a solution for 1000 cases, in this case your complexity is $$$10^8$$$, but multiplied by many constants such as functions parameter and body, and so on
Update
I made a mistake above: the code will never reach to 10000 because you have visited array
A small trick for Problem F
If you represent {DR, DL, UR, UL} as {0, 1, 2, 3} and have $$$d$$$ as direction variable, you can change directions easily
I dont know why is this crazy. The same solution worked if I made the following changes.
https://mirror.codeforces.com/contest/1807/submission/198324227
I think you just decreased the constant value
You can check this solution if you need a simplified one 198332947
vis = vector<vector<vector>>(n, vector<vector>(m, vector(4))); Sir this is a crime
Are ratings changed yet?
It doesn't changes individually.Either it will change for all or none.
когда будет системное тестирование?
Rating should be update before today's Div 2 round start.
Why are they redoing system testing?
I have a question related to Problem D. Odd Queries If the queries affect future queries. What would be the solution for this
Probably Segment-Tree or other similar DS's.
doubt segment tree because it does point update in log(N) but here we are changing entire subsegment (L to R with k) right? so time complexity per update will still be N*log(N)
thus time complexity to process all queries will be Q*N*log(N)
segment tree can do range update in O(logN) with lazy propagation
my bad i assumed lazy propagation would imply we need to check for the condition in the end i.e., check if the updated range sum is odd or not , i learnt something new thank you :)
I was Thinking Like You
I solved the problem D. In the contest, but After re-System Testing i get TLE :(
"Regardless of whether you are a trusted participant of the fourth division or not, if your rating is less than 1400 (or you are a newcomer/unrated), then the round will be rated for you."
My rating was 706 before attempting the contest, and it hasn't been affected yet. Wasn't the contest rated?
..
Solved the first 4 problems in 8 minutes and 5 problems in 17 minutes. Check the screen recording here https://youtu.be/HsGvOHmTwOw
why does problem 1807A - Plus or Minus have the "interactive" tag?