PROBLEM LINK:
Practice
Contest
Author: Pankaj Jindal, Piyush Kumar
Tester: Sergey Kulik
Editorialist: Aman jain
DIFFICULTY:
Medium
PREREQUISITES:
Set, Segment Tree
PROBLEM:
Given a array A of size N each filled with a color with value A_i. There are Q queries, each query update color of box at position Pos_i with color Col_i. For each query you need to find number of ways to select W adjacent boxes of same color from total of N boxes.
QUICK EXPLANATION:
Maintain a set of starting indices of color segments. Now on each query there are three possibilities, either we assign the same color, so no change to the array, or we have an update that breaks a continuous segment of colors, or an update that joins two continuous segments, each of which can be done in O(log(N)). Update your answer for the query and correspondingly update set of starting indices of color segments.
EXPLANATION:
Let’s start with easy constraints. Let say you have a color segment of length L, then number of ways to select W adjacent boxes from this segment would be max(0,(LW+1)). Now if you have sequence of segments say, s1s2s3s4, then number of ways to select W boxes would be sum of number of ways for each segment i.e. \sum{max(0,L_iW+1)}. Complexity of this solution would be O(N*Q), this works within timelimit for first subtask but will fail on main task.
To solve hard part efficiently, we can store the result of the query for the input string in ans and maintain set that store starting indices of color segments. After each query operation we need to insert or erase indices from set accordingly.
If you analyse closely you will find that each update operation looks like:
a) if update operation, does not change the state of boxes from current state, then there will be no change in answer to this query.
b) state change can be described as, ......s_a c1 s_b...... to ......s_a c2 s_b......, where s_a is segment to the left and s_b segment to the right of update position pos_i, with box color changed from c1 to c2.
Following cases possible (assuming 0 based indexing):

pos_i is 0 and c1 lies in s_b i.e. same in color to s_b, after the operation splitting takes place and ans = ans  ways(len(s_b)+1) + ways(len(1)) + ways(len(s_b)), here ways(l) means number of ways to get W adjacent boxes from segment of length l.

pos_i is 0 and c1 does not lies in s_b, after updation c2 lies in s_b, so merging takes place. ans = ans  ways(1)  ways(len(s_b)) + ways(len(s_b)+1).

pos_i is 0 and c1 and c2 both does not lie in s_b, no change in answer.

pos_i is N1 and c1 lies in s_a, after the operation splitting takes place and ans = ans  ways(len(s_a)+1) + ways(len(1)) + ways(len(s_a)).

pos_i is N1 and c1 does not lies in s_a, after updation c2 lies in s_a, so merging takes place. ans = ans  ways(1)  ways(len(s_a)) + ways(len(s_a)+1).

pos_i is N1 and c1 and c2 both does not lie in s_a, no change in answer.
 for rest of the cases pos_i lies in between 0 and N1, s_a = s_b, i.e. same in color and c1 lies in s_a, now splitting takes place, ans = ans  ways(len(s_a)+1+len(s_b)) + ways(len(s_a)) + ways(1) + ways(len(s_b)).

s_a = s_b, c1 does not lies in s_a, c2 lies in s_a, ans = ans  + ways(len(s_a))  ways(1)  ways(len(s_b)) + ways(len(s_a)+1+len(s_b)).

s_a = s_b, c1 and c2 does not lies in s_a, no change in answer.

s_a != s_b, c1 lies in s_a and c2 lies in s_b, ans = ans  ways(len(s_a)+1)+ways(len(s_a))ways(len(s_b))+ways(len(s_b)+1).

s_a != s_b, c1 lies in s_a and c2 does not lie in both, ans = ans  ways(len(s_a)+1) + ways(1) + ways((len(s_a)).

s_a != s_b, c1 lies in s_b and c2 lies in s_a, ans = ans  ways(len(s_b)+1)+ways(len(s_b))ways(len(s_a))+ways(len(s_a)+1).

s_a != s_b, c1 lies in s_b and c2 does not lie in both, ans = ans  ways(len(s_b)+1) + ways(1) + ways((len(s_b)).

s_a != s_b, c1 does not lies in both s_a and s_b, c2 lies in s_a, ans = ans  ways(1)ways(len(s_a))+ways(len(s_a)+1).

s_a != s_b, c1 does not lies in both s_a and s_b, c2 lies in s_b, ans = ans  ways(1)ways(len(s_b))+ways(len(s_b)+1).

s_a != s_b, c1 and c2 does not lies in both s_a and s_b, no change in answer.
COMPLEXITY:
Each query operation requires insertion, deletion of elements from set and answer update, all can be done in O(log(N)). For Q queries, overall complexity would be O(Q*log(N)).
ALITER:
Tester Sergey also suggested that above problem can be solved using Segment Tree, which would be much modular approach than handling many cases. HINT: For each node of segment tree stores length of color segment from left and length of color segment from right, color of left and right segment and ans for query for segment range. Update these values accordingly on each query operation.
If you are not able to solve the problem using Segment Tree, go through Sergey code in refernces.
AUTHOR'S, TESTER'S and Editorialist's SOLUTIONS:
author
tester
editorialist