We can use run-length encoding (i.e., RLE) to encode a sequence of integers. In a run-length encoded array of even length encoding
(0-indexed), for all even i
, encoding[i]
tells us the number of times that the non-negative integer value encoding[i + 1]
is repeated in the sequence.
arr = [8,8,8,5,5]
can be encoded to be encoding = [3,8,2,5]
. encoding = [3,8,0,9,2,5]
and encoding = [2,8,1,8,2,5]
are also valid RLE of arr
.Given a run-length encoded array, design an iterator that iterates through it.
Implement the RLEIterator
class:
RLEIterator(int[] encoded)
Initializes the object with the encoded array encoded
.int next(int n)
Exhausts the next n
elements and returns the last element exhausted in this way. If there is no element left to exhaust, return -1
instead.
Example 1:
Input ["RLEIterator", "next", "next", "next", "next"] [[[3, 8, 0, 9, 2, 5]], [2], [1], [1], [2]] Output [null, 8, 8, 5, -1] Explanation RLEIterator rLEIterator = new RLEIterator([3, 8, 0, 9, 2, 5]); // This maps to the sequence [8,8,8,5,5]. rLEIterator.next(2); // exhausts 2 terms of the sequence, returning 8. The remaining sequence is now [8, 5, 5]. rLEIterator.next(1); // exhausts 1 term of the sequence, returning 8. The remaining sequence is now [5, 5]. rLEIterator.next(1); // exhausts 1 term of the sequence, returning 5. The remaining sequence is now [5]. rLEIterator.next(2); // exhausts 2 terms, returning -1. This is because the first term exhausted was 5, but the second term did not exist. Since the last term exhausted does not exist, we return -1.
Constraints:
2 <= encoding.length <= 1000
encoding.length
is even.0 <= encoding[i] <= 109
1 <= n <= 109
1000
calls will be made to next
.This problem involves designing an iterator for a Run-Length Encoded (RLE) array. RLE is a data compression technique where consecutive runs of the same value are represented by a pair: (count, value). The RLEIterator
class needs to efficiently handle requests for the next n
elements from this compressed representation.
Approach:
The most efficient approach uses two pointers:
i
: Index to track the current (count, value) pair in the encoding
array. We increment i
by 2 to move to the next pair.j
: Index to keep track of how many elements have been consumed within the current pair.The next(n)
method iteratively consumes n
elements:
encoding[i] - j
) are sufficient to fulfill the request for n
elements.n
by subtracting the remaining elements of the current run and moves to the next pair by incrementing i
by 2 and resetting j
to 0. This process continues until either n
becomes 0 or we run out of pairs.j
by n
, effectively consuming n
elements, and returns the value associated with the current pair (encoding[i + 1]
).i >= encoding.length
) and still haven't fulfilled the request, it means there are no more elements, so it returns -1.Time and Space Complexity:
next(n)
is O(k), where k is the number of pairs we need to traverse before fulfilling the request for n elements. In the worst case, k could be proportional to the length of the encoding
array. However, it's amortized O(1) because each next
call consumes elements, and we don't repeatedly process the same pairs. The overall time complexity considering multiple next
calls is still bound by the number of elements processed (which is at most the sum of all counts in the encoding).Code Examples (Python):
class RLEIterator:
def __init__(self, encoding: List[int]):
self.encoding = encoding
self.i = 0
self.j = 0
def next(self, n: int) -> int:
while self.i < len(self.encoding):
if self.encoding[self.i] - self.j >= n:
self.j += n
return self.encoding[self.i + 1]
else:
n -= (self.encoding[self.i] - self.j)
self.i += 2
self.j = 0
return -1
The other language examples (Java, C++, Go, TypeScript) follow the same logic, just with syntax variations specific to each language. Refer to the detailed code in the original response for these implementations.