Merge pull request #217 from jerryderry/dp-theory-python
minimum distance dynamic programming in python
This commit is contained in:
commit
9b7993b3c8
39
python/41_dynamic_programming/min_dist.py
Normal file
39
python/41_dynamic_programming/min_dist.py
Normal file
@ -0,0 +1,39 @@
|
||||
"""
|
||||
Author: Wenru Dong
|
||||
"""
|
||||
|
||||
from typing import List
|
||||
from itertools import accumulate
|
||||
|
||||
def min_dist(weights: List[List[int]]) -> int:
|
||||
"""Find the minimum weight path from the weights matrix."""
|
||||
m, n = len(weights), len(weights[0])
|
||||
table = [[0] * n for _ in range(m)]
|
||||
# table[i][j] is the minimum distance (weight) when
|
||||
# there are i vertical moves and j horizontal moves
|
||||
# left.
|
||||
table[0] = list(accumulate(reversed(weights[-1])))
|
||||
for i, v in enumerate(accumulate(row[-1] for row in reversed(weights))):
|
||||
table[i][0] = v
|
||||
for i in range(1, m):
|
||||
for j in range(1, n):
|
||||
table[i][j] = weights[~i][~j] + min(table[i - 1][j], table[i][j - 1])
|
||||
return table[-1][-1]
|
||||
|
||||
|
||||
def min_dist_recur(weights: List[List[int]]) -> int:
|
||||
m, n = len(weights), len(weights[0])
|
||||
table = [[0] * n for _ in range(m)]
|
||||
def min_dist_to(i: int, j: int) -> int:
|
||||
if i == j == 0: return weights[0][0]
|
||||
if table[i][j]: return table[i][j]
|
||||
min_left = float("inf") if j - 1 < 0 else min_dist_to(i, j - 1)
|
||||
min_up = float("inf") if i - 1 < 0 else min_dist_to(i - 1, j)
|
||||
return weights[i][j] + min(min_left, min_up)
|
||||
return min_dist_to(m - 1, n - 1)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
weights = [[1, 3, 5, 9], [2, 1, 3, 4], [5, 2, 6, 7], [6, 8, 4, 3]]
|
||||
print(min_dist(weights))
|
||||
print(min_dist_recur(weights))
|
Loading…
Reference in New Issue
Block a user