mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-30 16:31:08 +00:00
192 lines
5.5 KiB
Python
192 lines
5.5 KiB
Python
|
"""
|
|||
|
Question:
|
|||
|
Given a binary matrix mat of size n * m, find out the maximum size square
|
|||
|
sub-matrix with all 1s.
|
|||
|
|
|||
|
---
|
|||
|
Example 1:
|
|||
|
|
|||
|
Input:
|
|||
|
n = 2, m = 2
|
|||
|
mat = [[1, 1],
|
|||
|
[1, 1]]
|
|||
|
|
|||
|
Output:
|
|||
|
2
|
|||
|
|
|||
|
Explanation: The maximum size of the square
|
|||
|
sub-matrix is 2. The matrix itself is the
|
|||
|
maximum sized sub-matrix in this case.
|
|||
|
---
|
|||
|
Example 2
|
|||
|
|
|||
|
Input:
|
|||
|
n = 2, m = 2
|
|||
|
mat = [[0, 0],
|
|||
|
[0, 0]]
|
|||
|
Output: 0
|
|||
|
|
|||
|
Explanation: There is no 1 in the matrix.
|
|||
|
|
|||
|
|
|||
|
Approach:
|
|||
|
We initialize another matrix (dp) with the same dimensions
|
|||
|
as the original one initialized with all 0’s.
|
|||
|
|
|||
|
dp_array(i,j) represents the side length of the maximum square whose
|
|||
|
bottom right corner is the cell with index (i,j) in the original matrix.
|
|||
|
|
|||
|
Starting from index (0,0), for every 1 found in the original matrix,
|
|||
|
we update the value of the current element as
|
|||
|
|
|||
|
dp_array(i,j)=dp_array(dp(i−1,j),dp_array(i−1,j−1),dp_array(i,j−1)) + 1.
|
|||
|
"""
|
|||
|
|
|||
|
|
|||
|
def largest_square_area_in_matrix_top_down_approch(
|
|||
|
rows: int, cols: int, mat: list[list[int]]
|
|||
|
) -> int:
|
|||
|
"""
|
|||
|
Function updates the largest_square_area[0], if recursive call found
|
|||
|
square with maximum area.
|
|||
|
|
|||
|
We aren't using dp_array here, so the time complexity would be exponential.
|
|||
|
|
|||
|
>>> largest_square_area_in_matrix_top_down_approch(2, 2, [[1,1], [1,1]])
|
|||
|
2
|
|||
|
>>> largest_square_area_in_matrix_top_down_approch(2, 2, [[0,0], [0,0]])
|
|||
|
0
|
|||
|
"""
|
|||
|
|
|||
|
def update_area_of_max_square(row: int, col: int) -> int:
|
|||
|
|
|||
|
# BASE CASE
|
|||
|
if row >= rows or col >= cols:
|
|||
|
return 0
|
|||
|
|
|||
|
right = update_area_of_max_square(row, col + 1)
|
|||
|
diagonal = update_area_of_max_square(row + 1, col + 1)
|
|||
|
down = update_area_of_max_square(row + 1, col)
|
|||
|
|
|||
|
if mat[row][col]:
|
|||
|
sub_problem_sol = 1 + min([right, diagonal, down])
|
|||
|
largest_square_area[0] = max(largest_square_area[0], sub_problem_sol)
|
|||
|
return sub_problem_sol
|
|||
|
else:
|
|||
|
return 0
|
|||
|
|
|||
|
largest_square_area = [0]
|
|||
|
update_area_of_max_square(0, 0)
|
|||
|
return largest_square_area[0]
|
|||
|
|
|||
|
|
|||
|
def largest_square_area_in_matrix_top_down_approch_with_dp(
|
|||
|
rows: int, cols: int, mat: list[list[int]]
|
|||
|
) -> int:
|
|||
|
"""
|
|||
|
Function updates the largest_square_area[0], if recursive call found
|
|||
|
square with maximum area.
|
|||
|
|
|||
|
We are using dp_array here, so the time complexity would be O(N^2).
|
|||
|
|
|||
|
>>> largest_square_area_in_matrix_top_down_approch_with_dp(2, 2, [[1,1], [1,1]])
|
|||
|
2
|
|||
|
>>> largest_square_area_in_matrix_top_down_approch_with_dp(2, 2, [[0,0], [0,0]])
|
|||
|
0
|
|||
|
"""
|
|||
|
|
|||
|
def update_area_of_max_square_using_dp_array(
|
|||
|
row: int, col: int, dp_array: list[list[int]]
|
|||
|
) -> int:
|
|||
|
if row >= rows or col >= cols:
|
|||
|
return 0
|
|||
|
if dp_array[row][col] != -1:
|
|||
|
return dp_array[row][col]
|
|||
|
|
|||
|
right = update_area_of_max_square_using_dp_array(row, col + 1, dp_array)
|
|||
|
diagonal = update_area_of_max_square_using_dp_array(row + 1, col + 1, dp_array)
|
|||
|
down = update_area_of_max_square_using_dp_array(row + 1, col, dp_array)
|
|||
|
|
|||
|
if mat[row][col]:
|
|||
|
sub_problem_sol = 1 + min([right, diagonal, down])
|
|||
|
largest_square_area[0] = max(largest_square_area[0], sub_problem_sol)
|
|||
|
dp_array[row][col] = sub_problem_sol
|
|||
|
return sub_problem_sol
|
|||
|
else:
|
|||
|
return 0
|
|||
|
|
|||
|
largest_square_area = [0]
|
|||
|
dp_array = [[-1] * cols for _ in range(rows)]
|
|||
|
update_area_of_max_square_using_dp_array(0, 0, dp_array)
|
|||
|
|
|||
|
return largest_square_area[0]
|
|||
|
|
|||
|
|
|||
|
def largest_square_area_in_matrix_bottom_up(
|
|||
|
rows: int, cols: int, mat: list[list[int]]
|
|||
|
) -> int:
|
|||
|
"""
|
|||
|
Function updates the largest_square_area, using bottom up approach.
|
|||
|
|
|||
|
>>> largest_square_area_in_matrix_bottom_up(2, 2, [[1,1], [1,1]])
|
|||
|
2
|
|||
|
>>> largest_square_area_in_matrix_bottom_up(2, 2, [[0,0], [0,0]])
|
|||
|
0
|
|||
|
|
|||
|
"""
|
|||
|
dp_array = [[0] * (cols + 1) for _ in range(rows + 1)]
|
|||
|
largest_square_area = 0
|
|||
|
for row in range(rows - 1, -1, -1):
|
|||
|
for col in range(cols - 1, -1, -1):
|
|||
|
|
|||
|
right = dp_array[row][col + 1]
|
|||
|
diagonal = dp_array[row + 1][col + 1]
|
|||
|
bottom = dp_array[row + 1][col]
|
|||
|
|
|||
|
if mat[row][col] == 1:
|
|||
|
dp_array[row][col] = 1 + min(right, diagonal, bottom)
|
|||
|
largest_square_area = max(dp_array[row][col], largest_square_area)
|
|||
|
else:
|
|||
|
dp_array[row][col] = 0
|
|||
|
|
|||
|
return largest_square_area
|
|||
|
|
|||
|
|
|||
|
def largest_square_area_in_matrix_bottom_up_space_optimization(
|
|||
|
rows: int, cols: int, mat: list[list[int]]
|
|||
|
) -> int:
|
|||
|
"""
|
|||
|
Function updates the largest_square_area, using bottom up
|
|||
|
approach. with space optimization.
|
|||
|
|
|||
|
>>> largest_square_area_in_matrix_bottom_up_space_optimization(2, 2, [[1,1], [1,1]])
|
|||
|
2
|
|||
|
>>> largest_square_area_in_matrix_bottom_up_space_optimization(2, 2, [[0,0], [0,0]])
|
|||
|
0
|
|||
|
"""
|
|||
|
current_row = [0] * (cols + 1)
|
|||
|
next_row = [0] * (cols + 1)
|
|||
|
largest_square_area = 0
|
|||
|
for row in range(rows - 1, -1, -1):
|
|||
|
for col in range(cols - 1, -1, -1):
|
|||
|
|
|||
|
right = current_row[col + 1]
|
|||
|
diagonal = next_row[col + 1]
|
|||
|
bottom = next_row[col]
|
|||
|
|
|||
|
if mat[row][col] == 1:
|
|||
|
current_row[col] = 1 + min(right, diagonal, bottom)
|
|||
|
largest_square_area = max(current_row[col], largest_square_area)
|
|||
|
else:
|
|||
|
current_row[col] = 0
|
|||
|
next_row = current_row
|
|||
|
|
|||
|
return largest_square_area
|
|||
|
|
|||
|
|
|||
|
if __name__ == "__main__":
|
|||
|
import doctest
|
|||
|
|
|||
|
doctest.testmod()
|
|||
|
print(largest_square_area_in_matrix_bottom_up(2, 2, [[1, 1], [1, 1]]))
|