Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

# | User | Rating |
---|---|---|

1 | MiFaFaOvO | 3520 |

2 | tourist | 3430 |

3 | apiadu | 3351 |

4 | mnbvmar | 3332 |

5 | Benq | 3290 |

6 | LHiC | 3276 |

7 | TLE | 3270 |

8 | Radewoosh | 3251 |

9 | ecnerwala | 3241 |

10 | Um_nik | 3240 |

# | User | Contrib. |
---|---|---|

1 | antontrygubO_o | 190 |

2 | Errichto | 188 |

3 | tourist | 179 |

4 | Radewoosh | 172 |

5 | pikmike | 166 |

6 | vovuh | 165 |

7 | ko_osaga | 162 |

8 | Um_nik | 161 |

9 | rng_58 | 155 |

10 | majk | 154 |

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial is loading...

Tutorial of Codeforces Round #612 (Div. 1)

Tutorial of Codeforces Round #612 (Div. 2)

↑

↓

Codeforces (c) Copyright 2010-2020 Mike Mirzayanov

The only programming contests Web 2.0 platform

Server time: Jan/19/2020 15:02:08 (e2).

Desktop version, switch to mobile version.

Supported by

User lists

Name |
---|

Could someone please explain the DP approach of problem C of Div-2 ?

there are two ways:

Firstly, let dp[i][j][k][0/1] means the answer to range 1-i with j odd numbers and k even numbers and the ith number is even/odd, so you can easily get the dp transition formula.

And the final answer is min(dp[n][num1][num0][0],dp[n][num1][num0][1]). num1 means the total odd number in range of 1-n and num0 is the even.（num1 equals to ceil(n/2) and num0 equals to floor(n/2) ) The time complexity is O(n³）

Secondly, since the number of even and odd used is sure in the range 1-i( if you use j even numbers, the number of odds is certainly i-j) , you can just use dp[i][j][0/1] where j means the number of evens in range 1-i (or that of odds if you like), and the dp dynamic formula will not change a lot. The time complexity is O(n²）

Can you explain the transition formula. I still can't get it.

My submission: 68264857. I did almost the same thing but by optimizing the recursion. What you do here is that if we have an empty slot we put a number that is divisible by 2 or not. Then, you have to compare the result of that recursion with out current result and print the final result.

Well, I gonna talk about the first type of dp transition formula and I am sure that you can think about the second one based on my explanations.

Firstly, based on the definition of the dp array, the formula is related to the value of ai. If ai ≠ 0， so we know whether it is odd or even. If it is an odd, the formula is " dp[i][j][k][1]=min(dp[i-1][j-1][k][0]+1, dp[i-1][j-1][k][1]) ". The explanation is that if you consider the value in the range 1~i with j odds and k evens used and the current one being odd, you need to think about the values, which you have calculated, in the range 1~(i-1), with (j-1) odds and k evens used since it is the only legal transition state to the current state. And whether the previous one you choose is odd or even depends on their values. Similarly, for a even number, it is " dp[i][j][k][0]=min(dp[i-1][j][k-1][0], dp[i-1][j][k-1][1]+1) ". Otherwise, ai is 0. So we dont exactly know whether it is odd or even, and we need to talk about both two situations. So both two above formulas should be considered.

Secondly, in order to get a correct value, we need to initialize the value of dp array correctly. So let's initialize them as 'inf'( a big value ). The reason is that 'inf' represents the illegal condition which we can avoid when we do 'min' operation. After that , just let dp[0][0][0][0]=dp[0][0][0][1]=0 because these two are legal situations.(Before anything starts, we stand at pos 0, use 0 odds and 0 evens, and the complexity is also 0).

So that it the first dp transition formula( the dp array of which consists many 'inf' values because there are many illegal situations ) , you can try to think about the second one, which consists less illegal situations.

Very nice explanation. Thank you.

Thank you for the explanation

`dp[i][j][k][1]=min(dp[i-1][j-1][k][0]+1, dp[i-1][j-1][k][1]) "`

Could you please tell me the significance of adding 1 to dp[i-1][j-1][k][0] in transition function ? Thanks in advance gold1103it is very simple. When you calculate dp[i-1][j-1][k][0],you put an even number on position i-1. So when you put an odd number(because the last index of the current condition is 1) on position i, the answer(the complexity according to the problem) should add 1

My submission based on your logic above, and which should be O(n^2), seems to TLE on case 57.

Can you take a look, please?

UPD: Nevermind! A stupid error when I was recomputing already computed values. This passes. Thanks!

Can you please help me in telling what is wrong with my approach I am using 3 state DP array Here is my code

we can do it without dp also solution without dp

count all odd space(where starting and ending is odd) and even space(where starting and ending is even) sort both odd and even array. put odd at odd place and even at even and check from starting and ending space also.

Time complexity:O(n).

This is not O(n)

Technically, it can be done in O(n) as the numbers are all from 1 to n using the counting sort.

Yes. Especially when the range is small.

quickest editorial !!

Could someone explain the approach for 1287D - Числа на дереве?

My solution involves some dp on subtrees.

Let's have a vector for every vertex $$$v$$$, where all vertices of its subtree are listed in increasing order of values assigned to them.

Then, the transition is not difficult. For a vertex $$$v$$$, glue together all vectors of its children and insert $$$v$$$ into it on the $$$c_v$$$-th position.

In order to get the answer, we look at the vector corresponding to the root of our tree: there, all vertices are listed in increasing order of values.

Submission: 68292734

Hey, your code seems to be really simple and neet. Can you please further explain it. I'm not able to understand your concept?

I didn't get it. Why is order[v] guaranteed to be listed in increasing order of values even before inserting the element in c[v]th position?

Congratulations on your elegant solution!

My solution of 1287D is just recursion:

68267302 Why it works? Well, because childs are independent, and parent value is selected to cut all child nodes at exact position.

Please someone answer why my code for div2 B 68280267 is getting TLE , though my complexity is O(n^2(k+log(n))

For this task, even PyPy doesn't help. You also using some set things that are not superfast. To pass the limit, you need to use some magic, like bitwise operations or other tricky things. Just look how other people solved it using python.

I did the same in c++ i.e using set to get the third character. set has an biggest constant that's why it is giving TLE (at least in c++ not sure in python). So u should try to change set to 3 if conditions.

for reference see this thread

But I used a set of only 3 elements :(

ya but still due to a huge constant associated with set, this makes your code slow.

https://codeforces.com/contest/1287/submission/68316351 this is O(k(n^2)log(n)) solution.But still getting TLE. please help me with this. thanks in advance.

I guess your solution fails because of sort in your next function. Replace it with simple if's and your solution should pass.

I changed it still getting TLE. https://codeforces.com/contest/1287/submission/68316468

Additional problem is that you performing temp=temp+symbol. You should do just temp+=symbol. Because temp=temp+symbol performs copying of string.

use (temp +=) instead of temp = temp + newchar when appending a char to a string

Hey can you look into my code too please

thank you so much.

use unordered_map instead of map .

Try hardcoding instead of the function nex, might help

Never try to access a value from map by

`operator[]`

if you're not sure it exists there. You should doOtherwise not existing value will be inserted with default constructor of int i.e. 0, which doesn't break the correctness but increases the time for the following tries to find a value

Why to use

`m.find(temp)`

if same can be done by using`m.count()`

function and it doesn't even return pointer it just return boolean value.That's because you need not only existence of the key, but its value as well. So in my case you do searching only once, in yours — twice at least

Wtf with the tags Div2A/Div2C???

1286 A — Garland (C of Div-2) could be solved in $$$O(n)$$$ if we use radix sort.

My "solution" for div 1C:

Query left and right half of string. Use basic backtracking to find all strings matching the answer, generate all candidates as concatenation of any possible string for left an right half. Calculate for each subsegment hash of answer for each string on this subsegment (can be done in O(n^2)) and find a subsegment giving distinct answers on all possible strings and query it. It takes two queries of length $$$n / 2$$$ and one of at most $$$n$$$ which is enough.

I managed to find 8 different strings with same answer, namely:

`abaabbabaabaabbabbabaabbab`

`abaabbabbabaabbabaabaabbab`

`abbabaabaabbabaabbabbabaab`

`abbabaabbabbabaabaabbabaab`

`baababbaabaababbabbaababba`

`baababbabbaababbaabaababba`

`babbaabaababbaababbabbaaba`

`babbaababbabbaabaababbaaba`

But in the official testcases my solution never encountered more than two candidates for each half.Any idea how do defeat it? 68322452

Could you replace the "1286B — Numbers on Tree" editorial's "Rightarrow" and "leq" with symbols. That would be appreciated. It appears only when the language is set to english.

Could you please attach the solution too?

Add solutions of those who contributed.

Can anyone please explain why the first code gets TLE on test 43 while the second one doesn't 68329420 68263645 DIV 2 Que B

Do yourself a favour and go see problem tags of div 2 C.

Could you give code for F? I get TLE with both fast zeta-transform 68330755 ($$$\mathcal{O}(2^{n} n^{2} \log n)$$$) and the $$$3^{n}$$$ approach 68330966, and AC only with fast randomised binary zeta-transform 68331478 ($$$\mathcal{O}(2^{n} nk \log n)$$$ where probability of success is polynomial in $$$(1 - 2^{-k})$$$).

(1 + sqrt(2))^n + n^3 * 2^nAnything else missing in the Div2 C taglist?

How come this is interactive ?? haha...

My O(nlogn) solution for Div2D/DIV1B Insert all numbers from 1 to n in a order-statistics tree. perform a dfs,every time you newly enter a node s,assign ans[s]=value at position c[s] in the order-statistics tree,and erase that element from order-statistics tree 68342107

It's actually not necessary to use an order-statistics tree here, my submission with $$$O(n^2)$$$ passes quite easily in 30 ms.

Can someone point out what am I missing in this DP approach to C-Garlands? https://codeforces.com/contest/1287/submission/68341470

There are few issues, first: if(dp[odd][even][level]!=-1) return dp[odd][even][level];

The above line doesn't make sense, cause if I am correct dp[odd][even][level] means there are "odd" number of odds and "even" number of even added to the the vector "a" till "level".

Imagine a vector, 0 2 0 4 5 0 Now, the answer is 1 2 6 4 5 3 that is 2 but your program will give 3 because of above statement.

According to your code, 1 2 6 4 is equivalent to 6 2 1 4, so if the program went on path 6 2 1 4 before 1 2 6 4, it will give the wrong answer. This is because the value of odd, even, the level is the same for both paths but since the path with greater value is calculated first, your dp[odd][even][level] may contain value which is not minimal for those value of odd, even and you are returning that non-minimal value without checking.

What you can do is remove that line and check all possible conditions but I tried it and it's resulting in TLE.

There's 1 more issue when you are setting a[level] to odd, you are not reverting it back to 0. I am no expert but that's all I could think of.

Can someone please help me with the editorial of LCC ? I have a few doubts.

How to calculate the probability that the $$$i$$$ th collision occurs first ? Why ? Can someone explain this ?

The editorial says that the answer for the mask is the probability that none of the $$$X$$$ collisions do not occur and the extremes move in the direction of the mask. What is $$$X$$$ here ?

Also, what do the leaves of the segment tree hold ? What is the meaning of adding a

ban?Please help me.

Even though I solved just now LCC and just checked isn't there better solution in editorial, I couldn't understand it in the first place. But, it turns out, it's describing same approach. So, not surprising that you didn't understand it.

Now, what I want to clarify is following. it says

`first i`

. It means first i collisions in list sorted by time. To make things clear, if there are n particles, then potentially you may have 3*n elements in this list. Each element of list is: which pair of two neighboring particles would collide in what configuration (three configuration described).Now, from where insight is comming. It's comming from neat fact that you may concatenate two segments using masks described. All you need to know, is probability to have ends of segment facing in certain directions, then you may concatenate them, and have new four values.

So, you represent whole array of particles as array of unit segments with 4 masks, and using segmented tree you may find probability of whole segment. The trick is, that you may change single unit segment to ban some of combinations, and recalculate probability of whole segment in log(n).

Finally, fact about occurance. When I realized it on my own, I was thinking in following way: lets calculate probability that nothing collide. It's just ban all collisions. Now, if we allow highest time of collision: there are two cases: nothing collide, or one or more collide at the same time the highest time of collision. So, probability that you have highest time collision is: probability of highest time of collision allowed

minusprobability of no collision allowed. Similarly, probability of second highest time is probability that it's allowed (and above)minusprobability of highest time collision allowed (and above).Thank you for your reply. If it is okay with you, can you please clarify some of my doubts ? Here is what I understood so far.

What I did not understand is how we use a segment tree to do this. Can you please help me ? :)

(1), (2) is right. (3) we don't change any directions of particles. We ban pairs of consecutive particles to collide in

exact configuration. So, for particles 5,6 we can doat once: LL — allow to collide, RR — ban, RL — allow for example (I don't care here about possibility of this case, but wecanban RR only). Segment tree is need to calculate probability of all this happen. If you have trouble with probability understanding, for n particles there is $$$100^n$$$ variations, and particle $$$i$$$ fly to the right in $$$p_i$$$ cases and to the left in $$$100-p_i$$$ cases, then you may count how many of configurations among $$$100^n$$$ is correct (taking bans into consideration), then divide the number by $$$100^n$$$ and it is probability. Then, apply expectation value formula by definition and you'll get answer.If you are free, can you please work out a small example $$$(n = 3)$$$ by hand and show how to do it on paper ? :)

Try to solve different problem. Statement is following: you're given $$$n,\;m,\; 0 \leq p_i \leq 10,\;0\leq t_i\leq 2,\;0\leq x_i\leq n-1$$$, it is number of particles, number of bans, probability 0 to 10, type of ban, and position of ban. You need to calculate number of arrays $$$a$$$ of length $$$n$$$ of numbers 0 to 9 such that some configurations banned.

Then, when you'll write this thing, find out how to add ban online and update answer fast.

Why is time limit getting exceeded on test case 10? I cannot understand. My submission https://codeforces.com/problemset/submission/1287/68273741

even i am getting TLE on test case 10 for 68349376

You can use combination and use map for count the same string. Here is my code

As it was mentioned in comment above, you should perform appending char to a string with S+=symbol, not S=S+symbol.

Thanks

it's interesting that the vast majority of div1 folks just decided to use dp in div1_a. i bet they knew about the solution the author presented here, but still they prefered dp. so the rule of thumb — avoid greedy approach if you are unsure about some parts of it?

I think it's more so that the DP solution is much cleaner to implement. With the greedy solution you also have to consider special cases (ie. intervals on the bounds of the array, $$$n=1$$$, all zeroes).

for problem B, why does 68349376 give TLE inspite of being n^2logn?

Your problem is in check function. Using set to get needed char increases complexity. And also you missed that fact that your algorithm is k*n^2*logn.

yeah,sorry for the complexity,but what should i do to optimise then?should i write using ifs?

yep, here's my submition of your algorithm with ifs that is AC https://codeforces.com/contest/1287/submission/68353515

my one also got accepted with ifs now 68356480 THANKS

Another approach is to note that it's possible to write a perfect hash function mapping a card to a 64-bit integer, since $$$3^k < 2^{64}$$$

Example: 68357302

Even more $$$4^k < 2^{64}$$$. So you can evaluate all symbols in parallel using bitwise operations. Add on top of that open hash table with square steps and you're blazing fast 68356690

I can't get it. Can somebody explain it please?

In 1286D, can 2 non-adjacent particles collide with each other for anyone else? Because, taking that assumption problem would be much easier.

Could you please show the solutions connected to these approaches?

Could someone explain the approach for doing div-1 question A though dp?

Wow, I have completely different solution to D. Mine is in fact much more complicated and fact that time of collision when both particles go facing each other is strictly less than time of collision when they go in the same direction is in fact crucial to my solution, while model solution just don't care. But I'm too lazy to describe it cause it's long x_0.

My (maybe simpler) solution for 1286D - LCC:

The main idea is still sorting all the collisions and calculate the probability that first $$$i$$$-th and $$$(i + 1)$$$-th will occur. When it comes to calculating these probability at some certain time, I have some following observations.

Lets first denotes the types of two protons colliding: $$$0$$$ — moving toward each other, $$$1$$$ — both moving to the left and $$$2$$$ — both moving to the right

Observations:

If we only take collisions of type $$$0$$$ into account, a consecutive segments where every protons have collisions with its neighbors, can only accept patterns like this: $$$«««...»»»$$$. Which means, we can always find a position which its left part will only move to the left and vice versa. Let's also denote this as special positions.

Events of type $$$1$$$ and $$$2$$$ will narrow the set of possible candidates for such positions in observation 1, but they still lie in a consecutive segment. It's easy to see that if some pair of $$$i$$$-th and $$$(i + 1)$$$-th protons can not both move to the left, we can not put such position anywhere in the right of them. So each reduction of these events is equal to banning a prefix or a suffix of possible candidates.

For a consecutive segment with the range of special positions, we can easily calculate the sum of probability that this segment will obey the mentioned pattern with prefix product array for moving to the left, suffix product for the right, and prefix sum of product of them (just some easy work on paper).

When we connect two components with some event of type $$$0$$$, it's a few casework to figure out the resulting component's range of special positions. And since events of type $$$1$$$ and $$$2$$$ always occur later than corresponding events of type $$$0$$$, two protons of these events always lie in the same components. So we just need to narrow the range and recalculate the probability.

My code: 68383975.

Hey,

I have some problems with understanding of the LCC problem.

Could you, please, briefly explain why in the third example the correct answer is 150902884?

To be honest, when I read the problem description I expected the solution to be a real number.

Well it's already stated in the statement that you need to print the answer as $$$P . Q^{-1}$$$ where $$$P/Q$$$ is the real number you mentioned.

Thanks for the reply.

But I am still confused. I was always taught that integers don't have an multiplicative inverse (besides -1 and 1). In addition, it is mentioned that $$$Q$$$ is a natural number, which does not have additive inverse as well. What does $$$Q^{-1}$$$ mean then?

Could you give an example? For instance, if the answer is 99/123, what should I output?

Here's how $$$123^{-1}$$$ modulo $$$998244353$$$ looks.

For $$$99/123$$$, just multiply them together and mod the result by $$$998244353$$$.

Thanks a lot! Now it makes sense!

Basically, ($$$123^{-1}$$$ mod 998244353) is 32463231, since 32463231*123 mod 998244353 equals to 1.

s = s + k creates a new string equal to s + k which takes $$$O(|s| + |k|)$$$ time and memory and then assigns it to s

s += k works in $$$O(|k|)$$$ (according to c++ standard might also work in $$$O(|s| + |k|)$$$, but that would be ridiculous to implement it this way)

Great, just learnt a better way of doing. Thanks.

Just notice you have Luftwaffe in your username. You still can change it for two days.

The Iran had bombed US bases and I have to prepare for word war 3.

So you can at least change it to 303 Squadron

Nopes, I don't trust brits.

Can someone point out what's wrong in my DP approach to C-Garlands? https://codeforces.com/contest/1287/submission/68427357

u need to keep track to previous element to. dp[i][o][e][flag] will be the right dp state.

Why? Can you please elaborate?

bcoz when a[i] != 0 and you are calling recurrence, and then when u r checking a[i] == 0 that a[i] != a[i-1] but how can a[i] be a[i-1] if for a[i] == 0 u r just making a[i] to 1 or a[i] to 2. you need to keep track of previous element whether it is odd or even or whether u filled it by even or odd. If you want code you can ping me.

Amritanshu_7

plz anyone tell me what is wrong in my approach of question B Hyperset 68457311

First: O(n^3k) has no chance of passing a maxtest

Second:

These three loops will run over some unordered triplets more than once.

can anyone point out the mistake for solution of problem-Garland my solution is this...

Looks like, even though according announcement 1286E - Гончар Фёдор наносит ответный удар will accept any answer by modulo 2^64, current checker wants full integer or there is some bug with comparison. unsigned long long didn't accepted for me, but bigint accepted.

if (a[i] % 2 || !a[i]) dp[i][j][1] = min(dp[i — 1][j][0] + 1, dp[i — 1][j][1]); if (a[i] % 2 == 0 && j) dp[i][j][0] = min(dp[i — 1][j — 1][0], dp[i — 1][j — 1][1] + 1);

The above code is a sanp from a garland problem, please can anyone explain what does i,j and 1 mean, i'm not able to get it with any comment

time complexity of question Dic2-C is O(n) not O(nlogn)

There is some sorting involved, that's why $$$O(n\log{n})$$$.

How to do OR convolution on independent subsets?

Can anyone please help me by telling that what am I doing wrong in my DP approach for problem C Garland.Click here to see my code.