I recently interviewed at Google. Because of this question my process didn't move forward after 2 rounds.
Suppose you are given an array of numbers. You can be given queries
to:
Find the sum of the values between indexes i and j.
Update value at index i to a new given value.
Find the maximum of the values between indexes i and j.
Check whether the subarray between indexes i and j, both inclusive, is in ascending or descending order.
I gave him a solution but it was to check the subarray between the indexes i and j. He asked me to optimize it. I thought of using a hashtable so that if the starting index is same and the ending index is more than the previous found, we store the maximum and whether its in ascending or descending and check only the remaining subarray. But that also didn't optimize it as much as required.
I'd love to know how I can optimize the solution so as to make it acceptable.
Constraints:
Everything from [1,10^5]
Thanks :)
All this queries can be answered in O(log N) time per query in the worst case(with O(N) time for preprocessing). You can just build a segment tree and maintain the sum, the maximum and two boolean flags(they indicate whether the range which corresponds to this node is sorted in ascending/descending order or not) for each node. All this values can be recomputed efficiently for an update query because only O(log N) nodes can change(they lie on the path from the root to a leaf which corresponds to the changing element). All other range queries(sum, max, sorted or not) are decomposed into O(log N) nodes(due to the properties of a segment tree), and it is easy to combine the value of two nodes in O(1)(for example, for sum the result of combining 2 nodes is just the sum of values for these nodes).
Here is some pseudo code. It shows what data should be stored in a node and how to combine values of 2 nodes:
class Node {
bool is_ascending
bool is_descending
int sum
int max
int leftPos
int rightPos
}
Node merge(Node left, Node right) {
res = Node()
res.leftPos = left.leftPos
res.rightPos = right.rightPos
res.sum = left.sum + right.sum
res.max = max(left.max, right.max)
res.is_ascending = left.is_ascending and right.is_ascending
and array[left.rightPos] < array[right.leftPos]
res.is_descending = left.is_descending and right.is_descending
and array[left.rightPos] > array[right.leftPos]
return res
}
As andy pointed out in the comments: The queries are quite different in nature, so the "best" solution will probably depend on which query type is executed most frequently.
However, the task
Find the sum of the values between indexes i and j.
can efficiently be solved by performing a scan/prefix sum computation of the array. Imagine an array of int values
index: 0 1 2 3 4 5 6
array: [ 3, 8, 10, -5, 2, 12, 7 ]
Then you compute the Prefix Sum:
index: 0 1 2 3 4 5 6, 7
prefix: [ 0, 3, 11, 21, 16, 18, 30, 37 ]
Note that this can be computed particularly efficient in parallel. In fact, this is one important building block of many parallel algorithms, as described in the thesis "Vector Models for Data-Parallel Computing" by Guy E. Blelloch (thesis as PDF File).
Additionally, it can be computed in two ways: Either starting with the value from array[0], or starting with 0. This will, of course, affect how the resulting prefix array has to be accessed. Here, I started with 0, and made the resulting array one element longer than the input array. This may also be implemented differently, but in this case, it makes it easier to obey the array limits (although one would still have to clarify in which cases indices should be considered as inclusive or exclusive).
However, given this prefix sum array, one can compute the sum of elements between indices i and j in constant time, by simply subtracting the corresponding values of the prefix sum array:
sum(n=i..j)(array[n]) = (prefix[j+1] - prefix[i])
For example:
sum(n=2..5)(array[n]) = 10 + (-5) + 2 + 12 = 19
prefix[5+1] - prefix[2] = 30 - 11 = 19
For task 2,
Update value at index i to a new given value.
this would mean that the prefix sums would have to be updated. This could be done brute-force, in linear time, by just adding the difference of the old value and the new value to all prefix sums that appear after the modified element (but for this, also see the notes in the last section of this answer)
The tasks 3 and 4
Find the maximum of the values between indexes i and j.
Check whether the subarray between indexes i and j, both inclusive, is in ascending or descending order.
I could imagine that the maximum value could simply be tracked while building the prefix sums, as well as checking whether the values are only ascending or descending. However, when values are updated, this information would have to be re-computed.
In any case, there are some data structures that deal with prefix sums in particular. I think that a Fenwick tree might allow to implement some of the O(n) operations mentioned above in O(logn), but I have not yet looked at this in detail.
Related
Given an array A consisting of N elements. Our task is to find the maximal subarray sum after applying the following operation exactly once:
. Select any subarray and set all the elements in it to zero.
Eg:- array is -1 4 -1 2 then answer is 6 because we can choose -1 at index 2 as a subarray and make it 0. So the resultatnt array will be after applying the operation is : -1 4 0 2. Max sum subarray is 4+0+2 = 6.
My approach was to find start and end indexes of minimum sum subarray and make all elements as 0 of that subarray and after that find maximum sum subarray. But this approach is wrong.
Starting simple:
First, let us start with the part of the question: Finding the maximal subarray sum.
This can be done via dynamic programming:
a = [1, 2, 3, -2, 1, -6, 3, 2, -4, 1, 2, 3]
a = [-1, -1, 1, 2, 3, 4, -6, 1, 2, 3, 4]
def compute_max_sums(a):
res = []
currentSum = 0
for x in a:
if currentSum > 0:
res.append(x + currentSum)
currentSum += x
else:
res.append(x)
currentSum = x
return res
res = compute_max_sums(a)
print(res)
print(max(res))
Quick explanation: we iterate through the array. As long as the sum is non-negative, it is worth appending the whole block to the next number. If we dip below zero at any point, we discard whole "tail" sequence since it will not be profitable to keep it anymore and we start anew. At the end, we have an array, where j-th element is the maximal sum of a subarray i:j where 0 <= i <= j.
Rest is just the question of finding the maximal value in the array.
Back to the original question
Now that we solved the simplified version, it is time to look further. We can now select a subarray to be deleted to increase the maximal sum. The naive solution would be to try every possible subarray and to repeat the steps above. This would unfortunately take too long1. Fortunately, there is a way around this: we can think of the zeroes as a bridge between two maxima.
There is one more thing to address though - currently, when we have the j-th element, we only know that the tail is somewhere behind it so if we were to take maximum and 2nd biggest element from the array, it could happen that they would overlap which would be a problem since we would be counting some of the elements more than once.
Overlapping tails
How to mitigate this "overlapping tails" issue?
The solution is to compute everything once more, this time from the end to start. This gives us two arrays - one where j-th element has its tail i pointing towards the left end of the array(e.g. i <=j) and the other where the reverse is true. Now, if we take x from first array and y from second array we know that if index(x) < index(y) then their respective subarrays are non-overlapping.
We can now proceed to try every suitable x, y pair - there is O(n2) of them. However since we don't need any further computation as we already precomputed the values, this is the final complexity of the algorithm since the preparation cost us only O(n) and thus it doesn't impose any additional penalty.
Here be dragons
So far the stuff we did was rather straightforward. This following section is not that complex but there are going to be some moving parts. Time to brush up the max heaps:
Accessing the max is in constant time
Deleting any element is O(log(n)) if we have a reference to that element. (We can't find the element in O(log(n)). However if we know where it is, we can swap it with the last element of the heap, delete it, and bubble down the swapped element in O(log(n)).
Adding any element into the heap is O(log(n)) as well.
Building a heap can be done in O(n)
That being said, since we need to go from start to the end, we can build two heaps, one for each of our pre-computed arrays.
We will also need a helper array that will give us quick index -> element-in-heap access to get the delete in log(n).
The first heap will start empty - we are at the start of the array, the second one will start full - we have the whole array ready.
Now we can iterate over whole array. In each step i we:
Compare the max(heap1) + max(heap2) with our current best result to get the current maximum. O(1)
Add the i-th element from the first array into the first heap - O(log(n))
Remove the i-th indexed element from the second heap(this is why we have to keep the references in a helper array) - O(log(n))
The resulting complexity is O(n * log(n)).
Update:
Just a quick illustration of the O(n2) solution since OP nicely and politely asked. Man oh man, I'm not your bro.
Note 1: Getting the solution won't help you as much as figuring out the solution on your own.
Note 2: The fact that the following code gives the correct answer is not a proof of its correctness. While I'm fairly certain that my solution should work it is definitely worth looking into why it works(if it works) than looking at one example of it working.
input = [100, -50, -500, 2, 8, 13, -160, 5, -7, 100]
reverse_input = [x for x in reversed(input)]
max_sums = compute_max_sums(input)
rev_max_sums = [x for x in reversed(compute_max_sums(reverse_input))]
print(max_sums)
print(rev_max_sums)
current_max = 0
for i in range(len(max_sums)):
if i < len(max_sums) - 1:
for j in range(i + 1, len(rev_max_sums)):
if max_sums[i] + rev_max_sums[j] > current_max:
current_max = max_sums[i] + rev_max_sums[j]
print(current_max)
1 There are n possible beginnings, n possible ends and the complexity of the code we have is O(n) resulting in a complexity of O(n3). Not the end of the world, however it's not nice either.
I recently came through an interesting coding problem, which is as follows:
There are n boxes, let us assume this is an array of n boxes.
For each index i of this array, three values are given -
1.) Weight(i)
2.) Left(i)
3.) Right(i)
left(i) means - if weight[i] is chosen, we are not allowed to choose left[i] elements from the left of this ith element.
Similarly, right[i] means if arr[i] is chosen, we are not allowed to choose right[i] elements from the right of it.
Example :
Weight[2] = 5
Left[2] = 1
Right[2] = 3
Then, if I pick element at position 2, I get weight of 5 units. But, I cannot pick elements at position {1} (due to left constraint). And cannot pick elements at position {3,4,5} (due to right constraint).
Objective - We need to calculate the maximum sum of the weights we can pick.
Sample Test Case :-
**Input: **
5
2 0 3
4 0 0
3 2 0
7 2 1
9 2 0
**Output: **
13
Note - First column is weights, Second column is left constraints, Third column is right constraints
I used Dynamic Programming approach(similar to Longest Increasing Subsequence) to reach a O(n^2) solution. But, not able to think of a O(n*logn) solution. (n can be up to 10^5.)
I also tried to use priority queue, in which elements with lower value of (right[i] + i) are given higher priority(assigned higher priority to element with lower value of "i", in case primary key value is equal). But, it is also giving timeout error.
Any other approach for this? or any optimization in priority queue method? I can post both of my codes if needed.
Thanks.
One approach is to use a binary indexed tree to create a data structure that makes it easy to do two operations in O(logn) time each:
Insert number into an array
Find maximum in a given range
We will use this data structure to hold the maximum weight that can be achieved by selecting box i along with an optimal selection of boxes to the left.
The key is that we will only insert values into this data structure when we reach a point where the right constraint has been met.
To find the best value for box i, we need to find the maximum value in the data structure for all points up to location i-left[i], which can be done in O(logn).
The final algorithm is to loop over i=0..n-1 and for each i:
Compute result for box i by finding maximum in range 0..(i-left[i])
Schedule the result to be added when we reach location i+right[i]
Add any previously scheduled results into our data structure
The final result is the maximum value in the whole data structure.
Overall, the complexity is o(nlogn) because each value of i results in one lookup and one update operation.
We have an array as input to production.
R = [5, 2, 8, 3, 6, 9]
If ith input is chosen the output is sum of ith element, the max element whose index is less than i and the min element whose index is greater than i.
For example if I take 8, output would be 8+5+3=16.
Selected items cannot be selected again. So, if I select 8 the next array for next selection would look like R = [5, 2, 3, 6, 9]
What is the order to choose all inputs with maximum output in total? If possible, please send dynamic programming solutions.
I'll start the bidding with an O(n2n) solution . . .
There are a number of ambiguities in your description of the problem, that you have declined to address in comments. None of these ambiguities affects the runtime complexity of this solution, but they do affect implementation details of the solution, so the solution is necessarily somewhat of a sketch.
The solution is as follows:
Create an array results of 2n integers. Each array index i will denote a certain subsequence of the input, and results[i] will be the greatest sum that we can achieve starting with that subsequence.
A convenient way to manage the index-to-subsequence mapping is to represent the first element of the input using the least significant bit (the 1's place), the second element with the 2's place, etc.; so, for example, if our input is [5, 2, 8, 3, 6, 9], then the subsequence 5 2 8 would be represented as array index 0001112 = 7, meaning results[7]. (You can also start with the most significant bit — which is probably more intuitive — but then the implementation of that mapping is a little bit less convenient. Up to you.)
Then proceed in order, from subset #0 (the empty subset) up through subset #2n−1 (the full input), calculating each array-element by seeing how much we get if we select each possible element and add the corresponding previously-stored values. So, for example, to calculate results[7] (for the subsequence 5 2 8), we select the largest of these values:
results[6] plus how much we get if we select the 5
results[5] plus how much we get if we select the 2
results[3] plus how much we get if we select the 8
Now, it might seem like it should require O(n2) time to compute any given array-element, since there are n elements in the input that we could potentially select, and seeing how much we get if we do so requires examining all other elements (to find the maximum among prior elements and the minimum among later elements). However, we can actually do it in just O(n) time by first making a pass from right to left to record the minimal value that is later than each element of the input, and then proceeding from left to right to try each possible value. (Two O(n) passes add up to O(n).)
An important caveat: I suspect that the correct solution only ever involves, at each step, selecting either the rightmost or second-to-rightmost element. If so, then the above solution calculates many, many more values than an algorithm that took that into account. For example, the result at index 1110002 is clearly not relevant in that case. But I can't prove this suspicion, so I present the above O(n2n) solution as the fastest solution whose correctness I'm certain of.
(I'm assuming that the elements are nonnegative absent a suggestion to the contrary.)
Here's an O(n^2)-time algorithm based on ruakh's conjecture that there exists an optimal solution where every selection is from the rightmost two, which I prove below.
The states of the DP are (1) n, the number of elements remaining (2) k, the index of the rightmost element. We have a recurrence
OPT(n, k) = max(max(R(0), ..., R(n - 2)) + R(n - 1) + R(k) + OPT(n - 1, k),
max(R(0), ..., R(n - 1)) + R(k) + OPT(n - 1, n - 1)),
where the first line is when we take the second rightmost element, and the second line is when we take the rightmost. The empty max is zero. The base cases are
OPT(1, k) = R(k)
for all k.
Proof: the condition of choosing from the two rightmost elements is equivalent to the restriction that the element at index i (counting from zero) can be chosen only when at most i + 2 elements remain. We show by induction that there exists an optimal solution satisfying this condition for all i < j where j is the induction variable.
The base case is trivial, since every optimal solution satisfies the vacuous restriction for j = 0. In the inductive case, assume that there exists an optimal solution satisfying the restriction for all i < j. If j is chosen when there are more than j + 2 elements left, let's consider what happens if we defer that choice until there are exactly j + 2 elements left. None of the elements left of j are chosen in this interval by the inductive hypothesis, so they are irrelevant. Choosing the elements right of j can only be at least as profitable, since including j cannot decrease the max. Meanwhile, the set of elements left of j is the same at both times, and the set of the elements right of j is a subset at the later time as compared to the earlier time, so the min does not decrease. We conclude that this deferral does not affect the profitability of the solution.
My mind is blown since I began, last week, trying to sort an array of N elements by condition: the difference between 2 elements being always less or equal to the next 2 elements. For example:
Α[4] = { 10, 2, 7, 4}
It is possible to rearrange that array this way:
{2, 7, 10, 4} because (2 - 7 = -5) < (7 - 10 = -3) < (10 - 4 = 6)
{4, 10, 7, 2} because (4 - 10 = -6) < (10 - 7 = 3) < (7 - 2 = 5)
One solution I considered was just shuffling the array and checking each time if it agreed with the conditions, an efficient method for a small number of elements, but time consuming or even impossible for a larger number of elements.
Another was trying to move elements around the array with loops, hoping again to meet the requirements, but again this method is very time consuming and also sometimes not possible.
Trying to find an algorithm doesn't seem to have any result but there must be something.
Thank you very much in advance.
I normally don't just provide code, but this question intrigued me, so here's a brute-force solution, that might get you started.
The concept will always be slow because the individual elements in the list to be sorted are not independent of each other, so they cannot be sorted using traditional O(N log N) algorithms. However, the differences can be sorted that way, which simplifies checking for a solution, and permutations could be checked in parallel to speed up the processing.
import os,sys
import itertools
def is_diff_sorted(qa):
diffs = [qa[i] - qa[i+1] for i in range(len(qa)-1)]
for i in range(len(diffs)-1):
if diffs[i] > diffs[i+1]:
return False
return True
a = [2,4,7,10]
#a = [1,4,6,7,20]
a.sort()
for perm in itertools.permutations(a):
if is_diff_sorted(perm):
print "Solution:",str(a)
break
This condition is related to differentiation. The (negative) difference between neighbouring elements has to be steady or increasing with increasing index. Multiply the condition by -1 and you get
a[i+1]-a[i] => a[i+2]-a[i+1]
or
0 => (a[i+2]-a[i+1])- (a[i+1]-a[i])
So the 2nd derivative has to be 0 or negative, which is the same as having the first derivative stay the same or changing downwards, like e.g. portions of the upper half of a circle. That does not means that the first derivative itself has to start out positive or negative, just that it never change upward.
The problem algorithmically is that it can't be a simple sort, since you never compare just 2 elements of the list, you'll have to compare three at a time (i,i+1,i+2).
So the only thing you know apart from random permutations is given in Klas` answer (values first rising if at all, then falling if at all), but his is not a sufficient condition since you can have a positive 2nd derivative in his two sets (rising/falling).
So is there a solution much faster than the random shuffle? I can only think of the following argument (similar to Klas' answer). For a given vector the solution is more likely if you separate the data into a 1st segment that is rising or steady (not falling) and a 2nd that is falling or steady (not rising) and neither is empty. Likely an argument could be made that the two segments should have approximately equal size. The rising segment should have the data that are closer together and the falling segment should contain data that are further apart. So one could start with the mean, and look for data that are close to it, move them to the first set,then look for more widely spaced data and move them to the 2nd set. So a histogram might help.
[4 7 10 2] --> diff [ 3 3 -8] --> 2diff [ 0 -11]
Here is a solution based on backtracking algorithm.
Sort input array in non-increasing order.
Start dividing the array's values into two subsets: put the largest element to both subsets (this would be the "middle" element), then place second largest one into arbitrary subset.
Sequentially put the remaining elements to either subset. If this cannot be done without violating the "difference" condition, use other subset. If both subsets are not acceptable, rollback and change preceding decisions.
Reverse one of the arrays produced on step 3 and concatenate it with other array.
Below is Python implementation (it is not perfect, the worst defect is recursive implementation: while recursion is quite common for backtracking algorithms, this particular algorithm seems to work in linear time, and recursion is not good for very large input arrays).
def is_concave_end(a, x):
return a[-2] - a[-1] <= a[-1] - x
def append_element(sa, halves, labels, which, x):
labels.append(which)
halves[which].append(x)
if len(labels) == len(sa) or split_to_halves(sa, halves, labels):
return True
if which == 1 or not is_concave_end(halves[1], halves[0][-1]):
halves[which].pop()
labels.pop()
return False
labels[-1] = 1
halves[1].append(halves[0][-1])
halves[0].pop()
if split_to_halves(sa, halves, labels):
return True
halves[1].pop()
labels.pop()
def split_to_halves(sa, halves, labels):
x = sa[len(labels)]
if len(halves[0]) < 2 or is_concave_end(halves[0], x):
return append_element(sa, halves, labels, 0, x)
if is_concave_end(halves[1], x):
return append_element(sa, halves, labels, 1, x)
def make_concave(a):
sa = sorted(a, reverse = True)
halves = [[sa[0]], [sa[0], sa[1]]]
labels = [0, 1]
if split_to_halves(sa, halves, labels):
return list(reversed(halves[1][1:])) + halves[0]
print make_concave([10, 2, 7, 4])
It is not easy to produce a good data set to test this algorithm: plain set of random numbers either is too simple for this algorithm or does not have any solutions. Here I tried to generate a set that is "difficult enough" by mixing together two sorted lists, each satisfying the "difference" condition. Still this data set is processed in linear time. And I have no idea how to prepare any data set that would demonstrate more-than-linear time complexity of this algorithm...
Not that since the diffence should be ever-rising, any solution will have element first in rising order and then in falling order. The length of either of the two "suborders" may be 0, so a solution could consist of a strictly rising or strictly falling sequence.
The following algorithm will find any solutions:
Divide the set into two sets, A and B. Empty sets are allowed.
Sort A in rising order and B in falling order.
Concatenate the two sorted sets: AB
Check if you have a solution.
Do this for all possible divisions into A and B.
Expanding on the #roadrunner66 analysis, the solution is to take two smallest elements of the original array, and make them first and last in the target array; take two next smallest elements and make them second and next-to-last; keep going until all the elements are placed into the target. Notice that which one goes to the left, and which one to the right doesn't matter.
Sorting the original array facilitates the process (finding smallest elements becomes trivial), so the time complexity is O(n log n). The space complexity is O(n), because it requires a target array. I don't know off-hand if it is possible to do it in-place.
I have a question in algorithm design about arrays, which should be implement in C language.
Suppose that we have an array which has n elements. For simplicity n is power of '2' like 1, 2, 4, 8, 16 , etc. I want to separate this to 2 parts with (n/2) elements. Condition of separating is lowest absolute difference between sum of all elements in two arrays for example if I have this array (9,2,5,3,6,1,4,7) it will be separate to these arrays (9,5,1,3) and (6,7,4,2) . summation of first array's elements is 18 and the summation of second array's elements is 19 and the difference is 1 and these two arrays are the answer but two arrays like (9,5,4,2) and (7,6,3,1) isn't the answer because the difference of element summation is 4 and we have found 1 . so 4 isn't the minimum difference. How to solve this?
Thank you.
This is the Partition Problem, which is unfortunately NP-Hard.
However, since your numbers are integers, if they are relatively low, there is a pseudo polynomial O(W*n^2) solution using Dynamic Programming (where W is sum of all elements).
The idea is to create the DP matrix of size (W/2+1)*(n+1)*(n/2+1), based on the following recursive formula:
D(0,i,0) = true
D(0,i,k) = false k != 0
D(x,i,k) = false x < 0
D(x,0,k) = false x > 0
D(x,i,0) = false x > 0
D(x,i,k) = D(x,i-1,k) OR D(x-arr[i], i-1,k-1)
The above gives a 3d matrix, where each entry D(x,i,k) says if there is a subset containing exactly k elements, that sums to x, and uses the first i elements as candidates.
Once you have this matrix, you just need to find the highest x (that is smaller than SUM/2) such that D(x,n,n/2) = true
Later, you can get the relevant subset by going back on the table and "retracing" your choices at each step. This thread deals with how it is done on a very similar problem.
For small sets, there is also the alternative of a naive brute force solution, which basically splits the array to all possible halves ((2n)!/(n!*n!) of those), and picks the best one out of them.