?detaR tI sI

6 years ago, # |

-6

I wish high ratings for everyone.

→ Reply

MRITuhin

6 years ago, # ^ |

← Rev. 2 →

+10

I also wish, but, how it possible to have high ratings to everyone, as the top of the standing possibly have rating increment where the bottom have possibility of rating decrement. :P

→ Reply

6 years ago, # ^ |

-8

it can happen, do your best.

→ Reply

fastboot

6 years ago, # |

← Rev. 2 →

Contest after a while and interestingly contest numbers are in binary (1010,1011) (; )).

→ Reply

variance

6 years ago, # ^ |

And 1010 is 10 in binary :D (1011 is also 11 in binary, of course)

→ Reply

Mockingjay

6 years ago, # |

+71

Unfortunately, Natasha’s flight won’t pass that successfully for me, she may fly to the Moon instead of Mars :P

→ Reply

6 years ago, # ^ |

+32

I will be happy if I can get the plane to take off :P

→ Reply

wallcrawler

6 years ago, # |

+19

After a wait of 10 days, coders are back to pavilion!! High ratings to everyone :)

→ Reply

wery0

6 years ago, # |

+48

In 2080: Codefoces Round #4000.

→ Reply

illchangeitnextyear

6 years ago, # |

+23

at least we still have two more "Educational Codeforces Round" 48 & 49 next two month maybe, which number begins with the digit '4' , and the next Round will be through only the 351 round

but surely we will enjoy this competition :)

→ Reply

MazzForces

6 years ago, # |

← Rev. 2 →

+17

I want to read this comment before round #4000. There will come a time when all of us will be old/dead, but the sport of cp will be one of the biggest.

→ Reply

andreumat

6 years ago, # ^ |

+25

I don't think that we are all will be so old. Suppose that CF holds a round once at every 3 days, it takes (4000-500)*3=10500 days or just 29 years. Due to the recent russian government's reform, we won't even reach the age of retirement.

→ Reply

6 years ago, # ^ |

← Rev. 2 →

+19

Lol no, there are 60-70 cf rounds per year, not 1 per 3 days

→ Reply

isaf

6 years ago, # |

+76

Lol. isaf27 and peltorator.

→ Reply

andreumat

6 years ago, # ^ |

+91

Your nickname makes the situation better.

→ Reply

Ne0n

6 years ago, # |

+31

If one can solve all the problems, will he be hired by Elon Musk?

→ Reply

tzc___________________wk

6 years ago, # |

← Rev. 2 →

_kun_? Is It a mathforces?

→ Reply

Jungarr1k

6 years ago, # ^ |

Probably it is anime.

→ Reply

6 years ago, # |

← Rev. 2 →

-15

A lot of days(8) for div3 participants wait for this contest.

Expectations are so high ratings.

→ Reply

tzc___________________wk

6 years ago, # ^ |

+12

Yes,it's 11 days for div.1 coders,wish more div.1 contests.

→ Reply

farmersrice

6 years ago, # |

+26

"The main heroine of the competition is astronaut Natasha, Mars researcher. If you solve all problems, her flight will pass successfully."

Is this just a cover up for the "tragic malfunction" of Natasha's mission? Remember, you heard it here first.

→ Reply

xh286286

6 years ago, # |

← Rev. 2 →

+14

In chinese, '4' is ominous .

→ Reply

Gnay_Oahnauhz

6 years ago, # ^ |

-10

'4' means death in Chinese

so it's nirvana after this contest

→ Reply

6 years ago, # ^ |

← Rev. 4 →

To be specific, 四（sì, meaning 4) sounds like 死 (sǐ, meaning death) in Mandarin. If Mandarin word for death was actually "4" that would be pretty funny :)

Edit: Oh no I have 4 upvotes!

→ Reply

6 years ago, # |

Say goodbye to codeforces round 4XX and say hello to codeforces round 5XX .

A new start for codeforces!(Shouldn't send us rating as celebrate?)

→ Reply

dxymaster2002

6 years ago, # |

-37

?detaR tI sI

→ Reply

dorijanlendvaj

6 years ago, # |

+66

"the next codeforces round(**Div.1, Div.2 or Div.3**) which begins with the digit 4..." What if codeforces round #4000 will be Div.4?

→ Reply

stefdasca

6 years ago, # |

+31

Almost every comment has upvotes, that's something new for Codeforces

→ Reply

JoaoMauricio

6 years ago, # ^ |

+54

That's because Wael.Al.Jamal didn't comment yet

→ Reply

swapnilr

6 years ago, # |

-24

"Do not skip this contest, because the next Codeforces Round (Div.1, Div.2, or Div.3), which number begins with the digit '4', will be through the 3501 round!"

Can anyone explain in simple terms what this means?

→ Reply

radoslav11

6 years ago, # ^ |

+36

499 + 3501 = 4000

→ Reply

peltorator

6 years ago, # ^ |

-7

Listen to the lecture, please!

→ Reply

radoslav11

6 years ago, # ^ |

+10

→ Reply

DeeaX99

6 years ago, # |

-10

i'm very excited for contest , i wish best for all

→ Reply

XORier

6 years ago, # ^ |

← Rev. 2 →

Me too, some of the great setters and testers are on board, hopes are high!

→ Reply

up-and-down

6 years ago, # |

+71

If you upvote this comment, you will get a positive rating change in the contest!

→ Reply

radoslav11

6 years ago, # ^ |

+31

If you upvote my comment, you will get LGM after the contest.

→ Reply

Emil_PhysMath

6 years ago, # ^ |

+23

I think if I upvote your comment then you will get positive contribution change.

→ Reply

Emil_PhysMath

6 years ago, # ^ |

-18

But it doesn't cancel positive rating change!!!

→ Reply

NatInTheHat

6 years ago, # ^ |

← Rev. 2 →

up-and-down It worked ;)

→ Reply

6 years ago, # ^ |

You saved me, somehow I got + even though my C failed systests :)

→ Reply

lazyneuron

6 years ago, # |

Best of Luck Minna-san!

→ Reply

C137

6 years ago, # |

First of all thanks for letting us know there is an interactive problem, secondly can we know what problem it will be??? Div.2 C/D/E/F ???

→ Reply

6 years ago, # ^ |

+20

Author:you're asking too much.:)

→ Reply

6 years ago, # |

-20

Its confirmed three legends (jqdai0815 , fateice , halyavin ) are not in this contest. Because they are the tester of this round.

→ Reply

solvemproblr

6 years ago, # |

+18

When will we see announcement of Div.1+Div.2 combined round?

→ Reply

Wael.Al.Jamal

6 years ago, # |

← Rev. 2 →

+114

→ Reply

ilias_kanchon1

6 years ago, # ^ |

See the DARKSIDE of today's round here.

→ Reply

peltorator

6 years ago, # |

-36

** ** *****?

→ Reply

Arthurvtr

6 years ago, # ^ |

-11

***.

→ Reply

MijPeter

6 years ago, # |

-27

We need to postpone contes 5 min. and I might make it in time from work. Thanks

→ Reply

Adhami

6 years ago, # |

← Rev. 2 →

+38

Since halyavin is a tester of this round. I think it will have very strong pretest.I hope halyavin will not fail me :')

Ouch.. so many hacks :D

→ Reply

6 years ago, # ^ |

lol

→ Reply

fastboot

6 years ago, # ^ |

I think halyavin didn't test problem C.Many solutions failed system test.

→ Reply

6 years ago, # |

It's really pitty there is no live disscussion after round with scott_wu and ecnerwala. Competition is more interesting for 30% to me.

→ Reply

azgar

6 years ago, # |

Going to be my first official contest. Any tips, anyone?

→ Reply

Kerim.K

6 years ago, # ^ |

+37

Ask "Is it rated?" in the comments.

→ Reply

gurshaan

6 years ago, # |

I am submitting my solution for Div2B and I am getting the wrong answer for pretest 4. It,s displaying that my answer is 2 whereas when checking on my personal IDE I am the getting the right answer that is 3. I have submitted the solution in different versions of c++ and every time I am getting a different answer for the exact same code. Why is so?

→ Reply

6 years ago, # ^ |

Possibly your problem works differently on your pc and codeforces due to undefined behaviour or some bug in your code.

You can try using the "Run" tab to run custom code with custom input on codeforces system.

You can also try using "diagnostics" compiler to detect such bugs

→ Reply

dengyaotriangle

6 years ago, # |

In problem E: their emperor decided to take tax from every tourist who visited the planet. Well,then theoretically Natasha does not have to give any money :I

→ Reply

Medusa

6 years ago, # ^ |

-8

lame joke :/

→ Reply

MaxZubec

6 years ago, # |

+31

div 2 = div 3

→ Reply

Scandium

6 years ago, # |

Interactive problems don't like me :(

→ Reply

natofp

6 years ago, # |

In D wha do i have to do with this: wrong output format Unexpected end of file — int32 expected?????

→ Reply

6 years ago, # ^ |

Your program was terminated before getting '0' answer from rocket.

→ Reply

riela

6 years ago, # |

Did someone forgot to a[i]%=k before taking the gcd in problem c? T_T

→ Reply

tzuyu_chou

6 years ago, # ^ |

+27

Does it matter? The answer should still be the same right?

→ Reply

6 years ago, # ^ |

-7

not if you dont take gcd with k

→ Reply

6 years ago, # ^ |

+14

Then it's a different mistake. For the only number k - 1 you must find gcd and don't have to do modulo.

→ Reply

subhamastoo

6 years ago, # |

Solution for question B??

→ Reply

6 years ago, # ^ |

Div2B was harder than A,C,D combined.

I made an array of food supply sizes (how much of each food), sorted them in decreasing order, and then greedily assigned people to food in the following way:

(used[i] represents how many currently assigned to food type i)

add 1 person to food type i as long as size[i]/(used[i]+1) is the maximum

you add n times, to find this maximum is O(n) so its n^2

but n^4 even passes so any polynomial solution should work

sorry I can't prove this, we will see by systests

→ Reply

https://pastebin.com/PBm7nExi

6 years ago, # ^ |

It was not hard

→ Reply

6 years ago, # ^ |

what if more than one person is assigned to one type of food?

→ Reply

6 years ago, # ^ |

For every food, I add to sum, the number of people that can be assigned to this food. So in the end, sum is equal to the maximum number of people I can feed.

→ Reply

6 years ago, # ^ |

← Rev. 2 →

OHHHH IT WAS ACTUALLY SO SIMPLE

QAQ

dude i feel stupid, thanks for the soluion though ^0^

EDIT:

I just realize function of answer space is monotonic.

This means b,c,d can all be solved with binary search. Damn

→ Reply

6 years ago, # ^ |

B and C are easier without

→ Reply

6 years ago, # ^ |

I would definitely agree B without, but I think C could be considered "easier" if you don't want to think about edge cases or math.

→ Reply

ay2306

6 years ago, # ^ |

It is not hard. Let the answer be k. you have to find maximum k which satisfies: sum different food packet quantity/k is greater than equal to n Solution link

→ Reply

PopescuMihai

6 years ago, # ^ |

Binary search is faster Ans = 0 Left = 1 Right = max value from used vector

→ Reply

utsav_deep

6 years ago, # ^ |

No, Div 2 B was just an easy binary search problem (other solutions are also there). But contrary to that you can say today's Div 2 D and E were of level of normal Div 2 C.

→ Reply

fugazi

6 years ago, # ^ |

I made an array b[101], where b[i] contains the frequency of a[i]. Now iterating i from 100 to 1 check whether if the food will last for i days (if(sum of b[j]/i) is greater than n).

→ Reply

subhamastoo

6 years ago, # ^ |

Thanks :)

→ Reply

raja1999

6 years ago, # |

what is the hack for Div1 B?

→ Reply

dorijanlendvaj

6 years ago, # ^ |

10000 1 10000

→ Reply

Amaryllis

6 years ago, # ^ |

I was hacked because I forgot to take the index in the sequence modulo N in the binary search part. My guess is most other hacks were because of this too.

→ Reply

Single_Ready_To_Mingle

6 years ago, # ^ |

How to solve it ?

→ Reply

6 years ago, # ^ |

+11

Print out '1' n times to get the sequence

then binary search

n = 30 log2(m = 1e9) = 29.somethings

should run in 30+30 = 60 iterations

but the sequence can "wrap around" is n is low but m is high, so you mod it or use a base-1 equivalent to modding

→ Reply

davidzhou

6 years ago, # ^ |

← Rev. 2 →

I did put the modulo and passed pretests, but then got a runtime error on test 24 and noticed I wrote mod M instead... :(

→ Reply

tarzanjunior

6 years ago, # |

-11

The following code didn't pass the pretests for A: https://ideone.com/jXjQQE

→ Reply

tarzanjunior

6 years ago, # ^ |

Why didn't it pass pretests even though it shows correct output on ideone?

→ Reply

abhayps

6 years ago, # |

+51

Binary search round.

→ Reply

saihemanth9019

6 years ago, # |

How to solve Div2E?

→ Reply

Vshining

6 years ago, # ^ |

← Rev. 2 →

BFS
failed system tests :(

→ Reply

T1duS

6 years ago, # ^ |

← Rev. 2 →

+12

Take gcd of all numbers including k
let that number be x
print all multiples of x

→ Reply

6 years ago, # ^ |

take the gcd of all numbers and k, let it be a. The answer is k / a and numbers are 0a, 1a, 2a...(k - 1)a

→ Reply

cartman94

6 years ago, # ^ |

Can you explain the logic behind this ?

→ Reply

6 years ago, # ^ |

← Rev. 3 →

+22

Digit d can be good if there is a solution for equation a₁ * x₁ + a₂ * x₂ + ... + a_n * x_n - k * x₀ = d. It exists if d ~~divides~~ is divisible by the gcd of all numbers and k.

→ Reply

cartman94

6 years ago, # ^ |

Is there any proof for this ?

→ Reply

6 years ago, # ^ |

+11

It is well-known criteria for solving linear diophantine equations

→ Reply

mohamedeltair

6 years ago, # ^ |

@Numb, Do you mean if d is divisible by the gcd (not divides)?

→ Reply

https://imgur.com/a/rDUIgBo

6 years ago, # ^ |

ah yes

→ Reply

vivek_shah

6 years ago, # ^ |

← Rev. 2 →

UPD : Got it!

→ Reply

Amaryllis

6 years ago, # |

← Rev. 3 →

Can someone explain to me why this solution for E is wrong?

Green points are beginning and end of open interval, red are the points when it was closed. If there was a red point between the 2 green ones the information is INCORRECT so we are left with this case only, right?

(Please ignore my paint skills)

Edit: Realised I misunderstood the statement , oops.

→ Reply

AkshajK

6 years ago, # |

+16

is div1E just find the bounding box of guaranteed Open values, find the bounding box of guaranteed Closed values (extending this to infinity in the opposite direction as the open values) and then just testing each of the k points to see if they are guaranteed open/close?

→ Reply

tzuyu_chou

6 years ago, # ^ |

+12

Find bounding box of open points (ez)
Find bounding box for each of the k points with the open points (ez)
For each of the k points find the number of closed points in the point's associated bounding box (not ez, 100 lines of 3d sum queries)

If a point is already in the bounding box of the open points, then it is open for sure. If the bounding box for one of the k points contains a closed point, then that point has to be closed, otherwise it is unknown.

→ Reply

AkshajK

6 years ago, # ^ |

← Rev. 3 →

ohh okay its more complicated than i thought NVM sorry

edit: are 3d sum queries necessary? doesnt it suffice just to find bounding box of open points, bounding box of closed points (extended to infinity), and then check if each other point is within one of the bounding boxes?

→ Reply

tzuyu_chou

6 years ago, # ^ |

I don't think that the bounding box of closed points is really a "box", it would be really hard to maintain that structure and check if a point is contained in the bounding shape of the closed points.

→ Reply

AkshajK

6 years ago, # ^ |

oh right nvm thanks!

→ Reply

chemthan

6 years ago, # |

← Rev. 2 →

+16

E: I think "check if there exist a point in a 3D rectangular" appeared somewhere.

→ Reply

Wild_Hamster

6 years ago, # ^ |

it definitely appeared in 2D version.

→ Reply

xrisk

6 years ago, # |

Every question is an integer such that 1 <= y <= m. NOOOOOOOOOO missed the <= m :(((

→ Reply

hocky

6 years ago, # |

+87

→ Reply

300iq

6 years ago, # ^ |

+27

There are solutions without binary search on D2B,C :)

→ Reply

hocky

6 years ago, # ^ |

← Rev. 2 →

+13

I personally am using *linear search on B

→ Reply

quanpham0805

6 years ago, # ^ |

Yeb I solved C in O(n) time by calculating from the last place to first one.

By the way may I ask for B without BS?

→ Reply

T1duS

6 years ago, # ^ |

No need for binary search in B. Linear search works well in time

→ Reply

6 years ago, # ^ |

← Rev. 2 →

Linear search is good enough for B (since the limit is so small) and C can be solved with math :) I would also argue a random solution might have a very small chance of passing for D...

→ Reply

dlwocks31

6 years ago, # |

← Rev. 2 →

I had no idea how to solve Div2E within TL, and I just wrote a random randomized solution and luckily it passed pretest..? And I don't even know why...?? 40811678

UPD: Failed System Test:)

→ Reply

AnasAbbas

6 years ago, # ^ |

-11

your solution takes 997 ms out of 1000 ms

i'm sorry but your solution most probably will fail on system test

→ Reply

gardusig

6 years ago, # ^ |

while(clock() < 0.99*CLOCKS_PER_SEC)

→ Reply

6 years ago, # |

This was a great contest, really enjoyed the problems! I made a dumb mistake on the interactive problem (div2D) unfortunately (forgot that upper limit is "m", not 1e9)...

→ Reply

Zovube

6 years ago, # |

+19

In Div2D problem statement says : If your program's request is not a valid integer between - 2³¹ and 2³¹−1 (inclusive) without leading zeros, then you can get any verdict.

Does this mean that if I request numbers in this range i will get the correct response with respect to problem statement? I got WA1 7 times and wasted some time because of requesting -1 to find out when the rocket lies xD

→ Reply

6 years ago, # ^ |

If you request a -1, you will get -2. You can only request numbers from 1 to m.

→ Reply

Zovube

6 years ago, # ^ |

Yeah, but for me this sentence was quite misleading...

→ Reply

6 years ago, # ^ |

for me to, but I quickly understood that I can request 1 (not -1) to find when it lies )

→ Reply

GNAQ

6 years ago, # |

BinarySearc(h)es

→ Reply

Diegogrc

6 years ago, # |

What is the test case 14 in Div2F?

→ Reply

6 years ago, # |

+26

When you rush to solve div1 D in the last 20 min, speedcode it and are ready to submit it in the very last minute... then notice XOR was also part of the list of operations. feelsbadman

→ Reply

6 years ago, # |

← Rev. 2 →

In Problem D, I couldn't accept the input using Python 2.

My simple solution:

n, m = map(int, raw_input().split())
print 1
raw_input()

The verdict: Idleness limit exceeded.

Any help is appreciated.

→ Reply

6 years ago, # ^ |

didn't flush after print

→ Reply

6 years ago, # ^ |

'print 1' prints '1'. I don't think we can flush this again.

→ Reply

6 years ago, # ^ |

print write to a buffer

stdout.flush() in Python; flushs the buffer to the stdout and clears the buffer

→ Reply

6 years ago, # ^ |

From the problem statement:

You can get "Idleness limit exceeded" if you don't print anything or if you forget to flush the output.

To flush the output buffer you can use (after printing a query and end-of-line):
...
stdout.flush() in Python;

→ Reply

6 years ago, # ^ |

← Rev. 2 →

I think you are right. But I've never used the 'flush()' method. And that's I thought 'print' prints in real-time.

→ Reply

chanchancl

6 years ago, # ^ |

it will be better when solving next interactive problem. or you can read guide for interactive problem: https://codeforces.net/blog/entry/45307

→ Reply

6 years ago, # ^ |

Thanks a lot.

→ Reply

6 years ago, # ^ |

Did you flush your output?

It was mentioned a few times in the contest announcement and another few times in the problem statement.

→ Reply

6 years ago, # ^ |

← Rev. 2 →

My solution: https://codeforces.net/contest/1011/submission/40799411

→ Reply

TheOneYouWant

6 years ago, # |

Div 2D Bound was very nice.... log(base2)(10^9) = 29.89, very close to 30, which was required for the solution! [60 requests — 30 to know true/false]

→ Reply

6 years ago, # ^ |

I mistakenly thought initially that we have to print the answer in the end again (not exiting if answer is 0) so that becomes 61 queries :P

I have to be careful while reading the questions next time

→ Reply

hocky

6 years ago, # ^ |

I think we can slide the truth/lie test. It is certain that the x will be greater than or equal to current y. It will reduce the requests slightly.

→ Reply

6 years ago, # |

+32

Coding speed testing round.

→ Reply

chadarmod

6 years ago, # ^ |

Yeah i second that. Problems were rather simpler demanding faster implementation by User

→ Reply

SuhasPai3799

6 years ago, # |

http://codeforces.net/contest/1011/submission/40794998 what's wrong with this method?

→ Reply

adamant

6 years ago, # |

Can someone elaborate how does this submission: 40792127 output exactly 60 numbers on hackcase:

1000000000 30 123
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

It is clearly seen that participant runs non-interrupting cycle with n = 30 iterations, then another one non-interrupting cycle with 30 iterations and afterwards he outputs answer. Am I missing something?

→ Reply

zoomswk

6 years ago, # ^ |

"If the program reads 0, then the distance is correct and you must immediately terminate the program (for example, by calling exit(0)). If you ignore this, you can get any verdict, since your program will continue to read from the closed input stream."

That's probably why. I think the interactor determines that the program is correct before the 61st query is printed (so basically that query is ignored).

→ Reply

tieros

6 years ago, # |

Solved Div.2 C — Fly on paper but too panicked to code it. Codeforces contests not good for my heart. Let me go and find some extra slow competitions :P

→ Reply

Gnay_Oahnauhz

6 years ago, # ^ |

Marathon Round 2

→ Reply

des1997

6 years ago, # |

problem E was easier than D

→ Reply

6 years ago, # ^ |

how to do?

→ Reply

6 years ago, # ^ |

+14

take the gcd of all numbers and k, let it be a. The answer is k / a and numbers are 0a, 1a, 2a...(k - 1)a

→ Reply

6 years ago, # |

← Rev. 2 →

+112

I spent the whole contest on F, coded it in a file named A.cpp and submitted F.cpp in the last minute. Looks like it could have passed :/

→ Reply

6 years ago, # |

after seeing problem D in div2, .........................................................????? what?

→ Reply

vivek_shah

6 years ago, # |

Seems Test Case 24 in Div2 Problem A is giving WA to many guys!!

→ Reply

6 years ago, # |

Please, can someone explain me, or give me some resource for understanding why next thing in binary search gives TLE:

while(l < r - 0.0000001) DoSomthing();

→ Reply

comosus7

6 years ago, # ^ |

If I used 1e-7 it gave me TLE, changed to 1e-6 and it passed.

→ Reply

6 years ago, # ^ |

God……I used 1e-8 :(

→ Reply

6 years ago, # ^ |

+16

1) might depend on doSomething()

2) Maybe for big values of l and r the precision isn't good enough and the value (l + r) / 2 is equal to l and r for the computer. You should stop earlier for big answer (it's ok because of relative error).

→ Reply

6 years ago, # ^ |

I believe it's a second thing. Definately precision wasn't good and I got TLE...

Only surprise is that I was using the same binary search for 100 times and never got similar result :)

→ Reply

6 years ago, # ^ |

try long double ?

I got tles when using double . This makes me anxious and ruined the contest :(

→ Reply

6 years ago, # ^ |

At the end I have finished everything with loop and iteration 10000 times.

It ruined my contest too :(

→ Reply

6 years ago, # ^ |

Usually for binary searching on doubles, I just do something like:

for(int step = 0; step < 50; ++step) DoSomething();

(One can, of course, adjust the 50 according to requirements/time limit etc)

→ Reply

6 years ago, # ^ |

I think its a floating-point precision thing, especially as numbers get larger and you can't store that many decimal places as 0.0000001

But you can fix the iterations by log2(1e9) = 30 (maybe 100 to be safe? I changed my 100 to 1000 LOL)

The tutorial solution to voltage keepsake did the same, probably because precision of floating-point is a common issue with binary searching on doubles

→ Reply

zhupengfei

6 years ago, # ^ |

+11

I used something more safe (in my opinion). I multiplied m by 1e9 and just used regular integer binary search with bounds [1, 1e18]. And then I divide the answer by 1e9 before outputing it. It passed the system tests: 40789915

→ Reply

6 years ago, # ^ |

← Rev. 2 →

That's way smarter solution than mine, thank you for sharing.

→ Reply

Gassa

6 years ago, # ^ |

When you have the answer close to 1E9, the values like 1E9 and 1E9 - 0.0000001 can't be distinguished by the double data type.

Now, take some two values for l and r which are consecutive possibilities to be stored in a double. What will be their average, m = (l + r) * 0.5? It can turn into either l or r, no middle ground. After that, if the checking function picks the unlucky one of the l = m and r = m (50% probability), the loop goes forever.

→ Reply

6 years ago, # ^ |

is the fix taking 1E9 + 1 as the r value initially?

→ Reply

Gassa

6 years ago, # ^ |

No, that won't help at all. The comparison while (l < r - 0.0000001) is the culprit.

The best fix is to instead use for (int step = 0; step < STEPS; step++) for binary search with floating point. With STEPS being obviously sufficient (200 for example).

→ Reply

6 years ago, # ^ |

← Rev. 2 →

I am not sure why this happened. Maybe some other mistake I committed since I took STEPS as 200 and it failed. But on resubmitting it after contest passed.

Failed Passed

The actual solution is at bottom (top is template only)

→ Reply

Gassa

6 years ago, # ^ |

In your case, the answer is dependent on whether if(check(mid)) happens at least once. So yeah, your change helps, since otherwise, the check never actually runs for the exact upper bound.

Another way to fix it is to use a single ans = lo; after the loop: regardless of what happens inside the loop, after it finished, the bounds lo and hi are both close enough to the real answer.

→ Reply

6 years ago, # ^ |

Thanks for the help!

→ Reply

6 years ago, # ^ |

I always prefer for loops when binary search in real numbers.
You could've done for (i=0;i<200;i++) binary_search()

→ Reply

kinetic_chase

6 years ago, # ^ |

← Rev. 2 →

I guess this method works too for binary-search on doubles.

ll hi=1e18,lo=1;
while(lo<hi){
ll mid=(lo+hi)/2;
ld target_value=mid/1e9; //reduced with 9 decimal precision 
} 
cout<<lo/1e9;

→ Reply

6 years ago, # |

Did anybody else do a sorting for div2 b?

→ Reply

ilias_kanchon1

6 years ago, # |

← Rev. 2 →

See the DARKSIDE of today's round here.

→ Reply

6 years ago, # |

← Rev. 2 →

+59

Solves A, B, C, D, E. Goes away for 50 minutes. Comes back. Sees: "solved: 2 out of 6". Deletes Codeforces account.

→ Reply

6 years ago, # ^ |

-14

hint

→ Reply

6 years ago, # ^ |

← Rev. 2 →

I tested them a little bit. I still don't know why they failed. Can somebody help me? 40790922 40796015

→ Reply

6 years ago, # ^ |

why you don't test negative fuel after landing

→ Reply

6 years ago, # ^ |

Because the landing doesn't take place if there's no fuel. Would that make a difference?

→ Reply

6 years ago, # ^ |

nvm the condition is in the for loop

→ Reply

Erdenebayar

6 years ago, # ^ |

for your D's solution, you should modulo nr by n

→ Reply

6 years ago, # ^ |

Thank you. I'm so stupid.

→ Reply

Erdenebayar

6 years ago, # ^ |

I'm so wondering, for your solution of C, you should read again Input area of this problem

→ Reply

SPatrik

6 years ago, # ^ |

← Rev. 2 →

+10

For problem C, I made the same mistake. When you have a solution, which you know is enough, you output it with precision 8. This means, that your output can be smaller than the solution you got, because of precision, and that little change of fuel might won't be enough to land.

→ Reply

6 years ago, # ^ |

Thanks. This sucks :(

→ Reply

Shivendra_420

6 years ago, # ^ |

Talking about mistakes look at mine, in Div2C instead of writing,

cout<<setprecision(10)<<ans<<endl;

I wrote,

cout<<ans<<setprecision(10)<<endl;

This was the only difference between right and wrong answer.

→ Reply

6 years ago, # ^ |

-8

Today in TCO round I made a bug, but after submitting I stress-tested, found a mistake, resubmitted and got AC. It's funny that the best advice I can give (for TC contests) was downvoted xd

→ Reply

Xellos

6 years ago, # ^ |

It was downvoted because it's not applicable to this case and you wrote it as if the person in question didn't already follow it.

There is no "best" advice. There are many types of generally good advice; some apply in some situations, some in other ones. In general, testing your solutions is useless if you misunderstand the problem, if there's a catch you haven't ever encountered and don't realise, if your solution is too slow on a specific (strongly non-random) type of tests, etc etc. Meta-advice like "analyse common mistakes you make and ways to avoid them" is better when followed, but that's also harder to do. Practice solving problems more, pick the right problems to practice on, put effort into focusing during contests (if only I could...), learn to read&understand statements faster, learn to do that more carefully, practice coding without bugs, use libraries for standard shit, practice typing faster, try different strategies that play to your strengths, get tools to automate things like testing, write your own e.g. generators, stresstesting on thousands of tests ties into this, there's so much that could be considered good advice... and you can still miss an obscure special case.

→ Reply

fastboot

6 years ago, # |

+81

Problem C

→ Reply

000coool

6 years ago, # ^ |

+18

but it seems you became handsome afterwards

→ Reply

solvemproblr

6 years ago, # |

So many red submissions in the standings page :(

→ Reply

6 years ago, # ^ |

-38

Every normal CF round should be like this round. It is awesome!

→ Reply

hi_im_bad2

6 years ago, # |

+32

shitty pretests.

→ Reply

fmqjpt

6 years ago, # |

+25

very weak pretests, wow

→ Reply

MStrechen

6 years ago, # ^ |

-22

That's what I call pretests.

→ Reply

AkiLotus

6 years ago, # |

May anyone give me a brief overview of test 63 in problem Div1C/Div2E? :<

→ Reply

Link: http://codeforces.net/contest/1010/submission/40805461

6 years ago, # ^ |

+12

2 30
6 10

→ Reply

AkiLotus

6 years ago, # ^ |

← Rev. 2 →

Got how I failed my solution now.

Thank you, for being incredibly responsive as a setter! ;)

→ Reply

DonMichaelCorleone

6 years ago, # |

3 consecutive Binary Search problems for div2. I can't believe !

→ Reply

Hasnaine_

6 years ago, # ^ |

4 actually. If you consider D also. :)

→ Reply

DonMichaelCorleone

6 years ago, # ^ |

Actually I was talking about B, C, D. What other problem did I miss ?

→ Reply

TadijaSebez

6 years ago, # |

Can someone prove exact time complexity of my ac solution for div1C? I can prove that it works better than O(k*p(k)) where p(k) is number of primes less than k.

→ Reply

tfg

6 years ago, # ^ |

← Rev. 2 →

It works in at most O(k * d(k)) where d(k) is the number of primes in the prime factorization of k because every time you run the inner loop, you end up being able to generate every number that's i * gcd(every number counted so far) and that gcd can change at most d(k) times.

Edit: Actually, it looks like it works in O(n + k)... since the worst possible input for that would be k = power of 2 and the inner loop would run O(1 + 2 + 4 + ... + k) = O(k)

→ Reply

TadijaSebez

6 years ago, # ^ |

Thank you :D

→ Reply

yiqzq

6 years ago, # |

maybe there are some mistakes in problem c for example ans > 1e9,not put -1,but get accepted.

→ Reply

Uzumaki_Narutoo

6 years ago, # |

← Rev. 2 →

in problem C , I see the we should use a of the first plant first then for all remaining plants should use b first the a as landing will be before flying and the order is very important as the weight of fuel will update after each fly or land I mean some thing like this

bool can(double fuel){
    double temp=(m+fuel)/a[0];
    fuel-=temp ;
    for(int i=1;i<n;++i){
        temp=(m+fuel)/b[i];
        fuel-=temp ;
        temp=(m+fuel)/a[i];
        fuel-=temp ;
    }
    temp=(m+fuel)/b[0];
    return (f-temp)>=0 ;
}

but I found some solutions had got AC and they don't care about the order of flying and landing ,How ? thanks in advance ^_^ UPD: I got confused in reading that code ,there is no problem xDD

→ Reply

Boolean17

6 years ago, # |

One of the best contests lately :) Problems were pretty cool and interesting. Thank you, VladProg and other people who helped to invent problems.

→ Reply

JoaoMauricio

6 years ago, # |

+25

Is CF-Predictor broken? It is not working for me in this contest.

→ Reply

flygrounder

6 years ago, # ^ |

Exciting fact: contest was on CF platform, but Predictor's servers are down ;)

→ Reply

Adhami

6 years ago, # ^ |

IDK why, but it happens too many times when we need it.

→ Reply

codewiz

6 years ago, # |

← Rev. 2 →

Why is including the constraint i.e. weight of fuel <= 1000000000 for Div2 C, giving WA for test case 76??

→ Reply

6 years ago, # |

+127

I think that my solution to F (upsolved) is interesting:

It turns out that we want to calculate for each i from [1, n] in how many ways can we cut off subtrees of tree so there will be i vertices left. Just doing fft in each vertex isn't enough (caterpillar is a countertest).

Let's use HLD. I have a recursive function (let's say DFS) which takes highest vertex from some path and asks for vector of results for this subtree. Sum of sizes of answers will amortize to O(n * log(n)). For each vertex on heavy path let's use DFS on every son from outside the path. In this problem tree was binary, but my solution works for any tree.

We have many vectors of results and we want to multiply them with fft. I'm using huffman codes on sizes of vectors, it's the best way probably. So for each vertex on path we have vector of results if the whole tree would consist only of this vertex and its sons except the one on this same heavy path. Now we could iterate over "which vertex on this path is cut off" and take product of every polynomial (vector of results) which is higher on path than this vertex. But it's quadratic of course. So we can do divide and conquer on this heavy path. Recursive function should return product of all polynomials in its interval and sum of prefix-products of polynomials. These two polynomials are enough to get the answer.

I'm sure that it's O(n * log³(n)), but it's probably O(n * log²(n)) also.

Here's a link to code: 40812966

→ Reply

6 years ago, # ^ |

← Rev. 3 →

+67

Nice solution.

Here is what I did:

Store the generating function whose coefficients are number of connected subgraphs rooted at i of a given size, for a subtree, T_i implicitly as a tuple of (j, F_i(x), G_i(x)), meaning T_i(x) = T_j(x) * F_i(x) + G_i(x), where j is some node in i's subtree. F_i and G_i will have the same degree.

When merging two subtrees, first find the explicit form using fast multiplication for the smaller subtree, and then multiply with larger subtree's polynomial's implicit form.

Also, calculate T_i as an explicit polynomial whenever the degree of F exceeds some value k, else leave it in the implicit form.

Optimal value of k comes to be $\text{[math]}$ , giving an $\text{[math]}$ solution.

Here's the code 40812980

I wish I had submitted the code for the right problem during the contest though :(

→ Reply

6 years ago, # ^ |

+15

Wow, I'm not much into generating functions but I'll for sure take a look into it. I guess that after parsing this comment for a few minutes I'll get it as brilliant :D

→ Reply

6 years ago, # ^ |

+10

Nope, It's similar, just sqrt instead of d&c :/

→ Reply

6 years ago, # ^ |

+10

Yeah, its quite similar. It has nothing fancy to do with generating functions. I just call polynomials as generating functions, when coefficients are of interest.

→ Reply

Petr

6 years ago, # ^ |

← Rev. 2 →

+14

Wow, I did exactly the same during the contest (but without a good value of k). However, even with the best value of k I still get TLE on test 18. I guess this is hard to squeeze in Java :)

→ Reply

300iq

6 years ago, # ^ |

← Rev. 2 →

-18

Well, good optimized our solution is working in ~2s, so you are like 2x slower than author solution :)

→ Reply

6 years ago, # ^ |

-181

Nobody cares about your solution radewoosh. India has jatin yadav, who could have solved it within contest and be the only one to do so but couldnt due to turnout of unfortunate events and so a big F YOU on your face.

→ Reply

6 years ago, # ^ |

+57

Get a life

→ Reply

6 years ago, # ^ |

+39

Yea, cause everybody cares about hates of some Specialist :/

→ Reply

6 years ago, # ^ |

+29

Toxic people like you should not be allowed to be in this community. We are all here to learn, and it's awesome to see people sharing their solutions and ideas, enabling us to learn not one but multiple approaches to a single problem.

But there's always that one guy like you, who has to type some shit up and ruin the experience. You should realise you're in no position to talk shit about someone exponentially better than you or compare two people so out of your league.

→ Reply

6 years ago, # ^ |

-65

The last time i saw something drop as hard as your rating, twin towers were nowhere to be seen lol.

→ Reply

6 years ago, # ^ |

Typical keyboard warrior. You know you could spend this time actually learning something.

As far as my rating goes, wait for today's changes :)

→ Reply

6 years ago, # ^ |

-39

Bitch, you will still be blue. No need to be gung-ho about it. As far as my learning goes, i am much more learned than you bitch. So stop throwing stones at others when your own house is made of glass.

→ Reply

6 years ago, # ^ |

-6

Lol kiddo you have no idea. It will hardly take me 4-5 contests to get to Div 1 if I start participating seriously.

And I don't remember the last time I saw a well-learnt person swearing just to prove he's smarter

→ Reply

6 years ago, # ^ |

-28

Stop doing drugs noobie. I have not used a single swear word. Your username clearly suits your mental state.

" It will hardly take me 4-5 contests to get to Div 1 if I start participating seriously." ----Many specialists and borderline experts have said the same to me a thousand times. Stop with the excuses already.

→ Reply

6 years ago, # ^ |

Look brother, I have better things to do than to argue with you. I'm not giving excuses, as I never said that I want to be in Div 1. I said that just because you were questioning my ability for no apparent reason. And yes, what I said is true, it won't take me long to get to Div1 if I really wanted to, but I'm not going to bust my ass just to prove my point to idiots like you. If I do get the time, I will participate seriously, and maybe then we'll talk.

→ Reply

6 years ago, # ^ |

-23

Haha, you seem so mentally,emotionally as well as sexually frustrated with your life. Learn to chill a little. Anyways, Good day "Mr intelligent and capable". You are so great that it might take you 1 contest to be yellow. You are great sir hats off.

→ Reply

6 years ago, # ^ |

← Rev. 2 →

+12

You seem insecure and angry ushagal0000, I believe the comments such as "mentally, emotionally as well as sexually frustrated with your life" might reflect your current position, and if that is the case please do something about it and not just label others with your own securities. There is no need to be so mad on an online platform, so "Learn to chill a little."

And go to facebook if you enjoy this bullshit.

→ Reply

6 years ago, # ^ |

-46

I have apologized...so all ya dickheads can stop with the downvotes

→ Reply

6 years ago, # ^ |

Why do you think downvotes exist?

→ Reply

Xellos

6 years ago, # ^ |

In this problem tree was binary, but my solution works for any tree.

I didn't even notice this and wrote my solution (similar to yours, splitting a path into two with similar sizes and processing them recursively instead of Huffman coding) for a general tree. Of course, the only difference is doing a convolution of many arrays.

I'm sure that it's O(n * log3(n)), but it's probably O(n * log2(n)) also.

It should be $\text{[math]}$ , logs don't tend to disappear in amortised complexity. However, I was thinking: is there a way to do convolution+something more efficiently by doing FFT first, then doing the something and then doing inverse FFT? After all, if an array with size 2ⁿ is padded with zeroes to size 2^n + k, we only need to compute the first 2ⁿ elements of its FFT and each of the remaining 2^k - 1 blocks of 2ⁿ elements is just the first block multiplied by some power of the primitive root. Then we could use $\text{[math]}$ if that sum of element-wise products can be computed easily.

→ Reply

6 years ago, # |

+15

This round was one of my most epic failures ever.

In Div1A I combined two arrays into one array of size 2·n. But when checking if there are ones among them I checked values only to n, not to 2·n. It was unlucky that no one in my room noticed that — one could get +100 on hacking and I could have a chance to find this awful mistake...

But Div1B was even more epic. More than 7 year ago I chose standard realization for binary search working in $\text{[math]}$ iterations. There is realization which needs $\text{[math]}$ iterations, but for me it was harder to work with bound values (when coding l = mid + 1 or r = mid - 1 there is no knowledge in general about bounds, while when l = mid or r = mid, it looked easier for me to keep one of bounds always "bad" and the other "good"). I never thought about difference in one iteration... And this was just the case — I failed in test making 61 iterations instead of 60.

Nevertheless, thanks to authors for the contest! It taught me a lot of things :)

→ Reply

flygrounder

6 years ago, # ^ |

I think you problem with B is somewhere else: I have AC with l=m, r=m

→ Reply

6 years ago, # ^ |

← Rev. 2 →

Looks like the difference in our codes is that you used r = m + 1 and I used r = m. It means that in your case l < x, r > x, while in mine l < x, r ≥ x. So when the difference between l and r becomes 2 your code immediately finds answer (l + r) / 2 = r - 1, while mine requires to do one more iteration for checking r - 1 and r.

→ Reply

flygrounder

6 years ago, # ^ |

← Rev. 2 →

Misunderstood you for the first time. Now it makes sense

→ Reply

6 years ago, # ^ |

It's not misunderstanding :) It's real cause of my problem. In most of binary search problems you don't have answer in a form < / = / > , it's usually like < / ≥ (or ≤ / > ). And in the second case it's usual way use binary search with one strictly bound and second non-strictly (like l < x, r ≥ x). While in a case < / = / > it's possible to use strictly bounds and do one iteration less. The same result could be obtained with mid ± 1 approach. But I just used standard way of binary search implementation without thinking about ± 1 extra iteration, and it was mistake.

BTW, I chose a way to write binary search without mid ± 1 when I learned C++ std::sort instead of standard C qsort. The first uses false / true for comparator while the second uses -1 / 0 / 1 :)

→ Reply

MStrechen

6 years ago, # ^ |

$\text{[math]}$ ? Could you please tell me more about this, I have AC with such kind of bs.

Do you have any proofs (links?) that mid ± 1 binsearch work with $\text{[math]}$ operations?

→ Reply

6 years ago, # ^ |

Looks like it's the same case as flygrounder's (see above).

Well, maybe it's not strictly proof, but if you use mid ± 1, after each iteration size of binary search segment becomes not more than ⌊ n / 2⌋, and it means that after $\text{[math]}$ iterations it will be equal to 1. While without ± 1 it would become ⌈ n / 2⌉ which is $\text{[math]}$

→ Reply

Kaban-5

6 years ago, # ^ |

← Rev. 3 →

TL;DR: in the sense of actual problem, yes. But this does not actually depend on how you code binary search, it only depends on whether you are actually using all information that is given to you.

I'll try to clarify. There are two different things, actually, that influence how you write integer binary search (I suppose that you write binary search in classic "l-r" style instead of "binary lifting" style).

"Model" you choose (this is not standard terminology, I just will use this word for purposes of this comment). Model indicates what do numbers l and r from binary search actually mean. There are 4 possibilities: the number you are searching for lies in [l, r] range, in [l, r) range, in (l, r] range or in (l, r) range. In all of this models m can be choosen as $\text{[math]}$ and terminating conditions for this models are r - l ≤ 0, r - l ≤ 1, r - l ≤ 1 and r - l ≤ 2 respectively.
What type of information you can get by asking a query. Usually, when you use binary search to solve a problem, you are asking questions that can be answered "yes" or "no". In this problem, however, you are asking a question that has three answers: the number you are searching for is less than, equal to or bigger than your guess.

Now, you could, say, write the following C++ code in [l, r) model:

Code that loses information

// we are searching for number X in range [1, n]
int l = 1, r = n + 1;
while (r - l > 1)
{
    // we already know that X lies in range [l, r)
    int m = (l + r) / 2;
    char sign = ask(m); // ask(m) returns sign in comparision m ? X
    if (sign == '=') exit(0);
    if (sign == '>')
        r = m;          // m > X implies that X lies in range [l, m)
    else		// sign == '<'
        l = m;          // otherwise X lies in range [m, r)
}

Notice that we lose some information here when sign == '<'. Indeed, we know that X lies in range [m, r), but we know also that X is not equal to m (otherwise sign == '='). Binary search still does terminate, but in possibly bigger number of steps than it could. Writing l = m + 1 instead of l = m fixes the problem.

Now suppose that we are searching for X in range [1, n]. We could write code that uses all the information given for all 4 possible models in the following ways:

[l, r]

int l = 1, r = n;
while (r - l > 0)
{
    // we already know that X lies in range [l, r]
    int m = (l + r) / 2;
    char sign = ask(m); // ask(m) returns sign in comparision m ? X
    if (m == '=') exit(0);
    if (m == '>')
        r = m - 1;      // m > X implies that X lies in range [l, m - 1]
    else                // sign == '<'
        l = m + 1;      // m < X implies X lies in range [m + 1, r]
}

[l, r)

int l = 1, r = n + 1;
while (r - l > 1)
{
    // we already know that X lies in range [l, r)
    int m = (l + r) / 2;
    char sign = ask(m); // ask(m) returns sign in comparision m ? X
    if (m == '=') exit(0);
    if (m == '>')
        r = m;      // m > X implies that X lies in range [l, m)
    else            // sign == '<'
        l = m + 1;  // m < X implies X lies in range [m + 1, r)
}

(l, r]

int l = 0, r = n;
while (r - l > 1)
{
    // we already know that X lies in range (l, r]
    int m = (l + r) / 2;
    char sign = ask(m); // ask(m) returns sign in comparision m ? X
    if (m == '=') exit(0);
    if (m == '>')
        r = m - 1;  // m > X implies that X lies in range (l, m - 1]
    else            // sign == '<'
        l = m;      // m < X implies X lies in range (m, r]
}

(l, r)

int l = 0, r = n + 1;
while (r - l > 1)
{
    // we already know that X lies in range (l, r)
    int m = (l + r) / 2;
    char sign = ask(m); // ask(m) returns sign in comparision m ? X
    if (m == '=') exit(0);
    if (m == '>')
        r = m;      // m > X implies that X lies in range (l, m)
    else            // sign == '<'
        l = m;      // m < X implies X lies in range (m, r)
}

Now, any of this codes indeed needs $\text{[math]}$ questions to figure out value of X for n = 2^k + 1 - 1 (and not more than that for smaller n's). In this exact problem, you still need to ask this number in the end.

Quick summary: number of queries does not depend on model you choose, because this choice is purely estetical. However, it does depend on whether you use all information given to you or not.

→ Reply

6 years ago, # ^ |

Nice intuitive explanation. This article also explains how to choose the model nicely.

→ Reply

6 years ago, # ^ |

ceil(log₂n) should have done fine, since log₂10⁹ ≈ 29.89.

→ Reply

6 years ago, # ^ |

Sorry, I counted only binary search iterations, without first n iterations and one extra iteration to print answer (if it wasn't already found).

→ Reply

Xellos

6 years ago, # ^ |

← Rev. 2 →

I also failed on 61 iterations, it seems. I understand the intent behind 60 queries was to make only $\text{[math]}$ algorithms pass, but I would decide against such a strict constraint if I were a problemsetter, because you can't reasonably separate solutions that can make one more query occasionally. Here's why: [submission:http://codeforces.net/contest/1010/submission/40792452] [submission:http://codeforces.net/contest/1010/submission/40895948]

Note the difference is purely cosmetic — it has no bearing on how the algorithm works, since decreasing the range of million possibilities by 30 is useless in that sense. It can only slightly change the numbers in some queries. I really didn't expect the out of contest submission to pass, it was just a "lol what if I wrote this instead", and yet it got AC. That idea isn't even something esoteric, not asking for the same number over and over can be a natural way someone would implement this algorithm.

Good test data should be made in such a way that intended solutions almost certainly pass, solutions intended to fail almost certainly fail and anything in between is irrelevant. In this case, solutions that can use 61 queries seem to be intended to fail (otherwise, why use 60 and not e.g. 64), but I found one that passes purely by accident, which is... really bad. My reaction to this as the problemsetter would be "oh shit, this can't happen".

→ Reply

Sharon

6 years ago, # |

+22

Top 4 in Div2 is just smurfs :(

→ Reply

peach

6 years ago, # ^ |

+16

→ Reply

6 years ago, # |

← Rev. 2 →

+13

Div2 C has an O(n) solution as well.
Let

$\text{[math]}$

where $A$ and B are the two lists given.
The answer is ans = x * (a - 1), where x is the initial weight of rocket.
Of course, the answer is 0 when any element in any of the two lists is 1.

→ Reply

quanpham0805

6 years ago, # |

Sadly none of Div1 participants suceeded in helping Natasha to reach Mars :(

Btw the problems are nice and tough especially the interactive, and for me the statement is a little bit hard to understand, maybe due to my English :(

→ Reply

HansvinTS

6 years ago, # |

-18

Hello VladProg, I found something wrong with the checker on Fly's problem. I got accepted instead of wrong answer on that problem. Or you could check on my submission for that problem. I hope you could fix it, so it will not affect any future contest :)

→ Reply

fmqjpt

6 years ago, # ^ |

+11

relative error is still less than 10^- 6 so it is still correct

→ Reply

HansvinTS

6 years ago, # ^ |

I don't think it is less than 10^-6

→ Reply

fmqjpt

6 years ago, # ^ |

+11

Relative error is (difference)/(answer) not just absolute difference

→ Reply

HansvinTS

6 years ago, # ^ |

Oh i see, i didn't read that statement carefully. Thank for clarifying it.

→ Reply

6 years ago, # ^ |

$\text{[math]}$

→ Reply

pankajAhuja

6 years ago, # |

In div2C can someone help me understand, why 40816252 works and 40815743 doesn't.

→ Reply

test616.cpp

6 years ago, # ^ |

Due to precision issues, when mid is very close to the answer, possible() might not return the expected value. Here the answer is exactly 10^9 but your possible() returned false so you printf("-1\n") and got WA. And the first one just happened to work.

→ Reply

xelwood

6 years ago, # ^ |

← Rev. 2 →

I have a code 40800183 that checks first if 10^9 is ok. It returns false. Then, there is no sense to do binary search after that. But why it does return false ? Maybe it is not 100% correct to require 10^9 answer in this case ? Because it seems to be something not well-defined distinction between -1 and 10^9 if there is less than 10^-6 difference.

→ Reply

pankajAhuja

6 years ago, # ^ |

from test616.cpp's explanation if the answer is equal to the upper bound, it may fail due to precision issues. However this is not a very good explanation, but from 40807764 you can see if you increase the search space significantly greater than the max possible answer you will get correct answers even if the answer is 10^9. In my case the precision issue was probably caused due to many calculations in one single line and yours it's probably because of double/int (which shouldn't be a problem). But maybe someone more informed can help us.

→ Reply

DeeaX99

6 years ago, # |

Binary search contest

→ Reply

timus123

6 years ago, # |

can someone explain me why this solution for D is giving wrong answer on pretest 1??

Link to solution : https://ideone.com/klzLiI

→ Reply

MStrechen

6 years ago, # ^ |

Sometimes you know the answer (hi — lo == 1), but didn't tell the system that answer (because you do nothing after while loop). It's just a brief note, got the same problem, I may be wrong.

→ Reply

JoaoMauricio

6 years ago, # |

+11

you know you're a very anxious person when you can't do anything while wait the rating change

→ Reply

natofp

6 years ago, # |

What's wrong with this code? Please take a look it is pretty short and clear : https://codeforces.net/contest/1010/submission/40818298

→ Reply

6 years ago, # ^ |

try while(a<b)

→ Reply

natofp

6 years ago, # ^ |

xDDDD Thanks.

→ Reply

no_1999

6 years ago, # |

-21

I got an email about my submission (40806184) for problem 1011A coincides with submission (40800257) of user 641999. Actually both accounts are mine. You can verify it with email id and name on account. I accidentally submitted solution in account 641999. After realising mistake i submitted solution in account no_1999. All submissions after that are from account no_1999.

→ Reply

6 years ago, # ^ |

+44

I think its not allowed to have multiple accounts.

→ Reply

peach

6 years ago, # ^ |

-8

Indeed you agree when registering for contest.

→ Reply

VLamarca

6 years ago, # |

The top5 from div2 is wrong as 2 of them probably were caught cheating

→ Reply

6 years ago, # ^ |

Thank you. Fixed.

→ Reply

peach

6 years ago, # |

OP, consider updating Div2 winners?

→ Reply

6 years ago, # ^ |

Yes. Fixed.

→ Reply

http://codeforces.net/contest/1011/submission/40802960

6 years ago, # |

← Rev. 4 →

A question about gcd in problem D.

Why would you always need to gcd K?

For example, if all n numbers followed this property x = (x mod k) then shouldn't it be unnecessary to consider k in your gcd (because it is no longer part of the equation)?

I submitted a code with something like

if (x != x%k) d = gcd(d, k);

and it gets wrong answer.

But when I just do d = gcd(d, k); it will work.

Why is that?

UPD:

Ok I got it thanks to Svlad_Cjelli (both smooth with algorithm and with the ladies) Basically if you consider factor x that exceeds K it will become x mod K, actually x mod k can be some factor you didn't consider before. Therefore, its not x % k = x where x, but all the time we mod k (because we can always add one item infinitely until it becomes greater or equal to k)

→ Reply

arnold518

6 years ago, # |

I approached Div2 Prob b in dynamic but i got TLE in TestCase 14 Since the n is so small, it should hold up to N^2 or even N^3 Why did i get TLE???

→ Reply