1 回答
TA贡献1847条经验 获得超11个赞
我已经实现了三种算法。
第一个算法是Simple
,使用最简单的嵌套循环方法,它具有O(N^5)
时间复杂度(对于我们的情况N
,其中 是输入网格的一侧10
),对于我们的输入大小10x10
时间来说O(10^5)
是相当好的。代码中的算法 ID 是algo = 0
。如果您只想查看该算法,请跳转到------ Simple Algorithm
代码内的行。
第二种算法是Advanced
,采用动态规划方法,其复杂度O(N^3)
比第一种算法快得多。代码中的算法 ID 是algo = 1
。跳转到------- Advanced Algorithm
代码内的行。
我实现的第三个算法Simple-ListComp
只是为了好玩,它几乎与 相同Simple
,相同的O(N^5)
复杂性,但使用 Python 的列表理解而不是常规循环,这就是为什么它更短,也慢一点,因为没有使用一些优化。代码中的算法 ID 是algo = 2
。跳转到------- Simple-ListComp Algorithm
代码内的行以查看算法。
除了算法之外,其余代码还实现检查结果的正确性(算法之间的双重检查)、打印结果、生成文本输入。代码分为解决任务函数solve()
和测试函数test()
。solve()
函数有许多参数来允许配置函数的行为。
所有主要代码行均由注释记录,阅读它们以了解如何使用代码。基本上,如果s
变量包含带有网格元素的多行文本,就像您的问题一样,您只需运行solve(s, text = True)
它就会解决任务并打印结果。algo = 0, check = False
此外,您还可以通过给出求解函数的下一个参数(此处 0 表示算法 0 )从两个版本(0(简单)和 1(高级)和 2(简单列表计算))中选择算法。查看test()
函数体以查看最简单的用法示例。
算法默认输出到控制台的所有簇,从最大到最小,最大的用.
符号表示,其余的用B
, C
, D
, ...,Z
符号表示。show_non_max = False
如果您只想显示第一个(最大)簇,您可以在求解函数中设置参数。
我将解释简单算法:
基本上算法的作用是 - 它搜索所有可能的有角度的
1s
矩形并将其中最大的矩形的信息存储到ma
二维数组中。Top-left
这个矩形的点是(i, j)
,top-right
-(i, k)
,bottom-left
-(l, j + angle_offset)
,bottom-right
-(l, k + angle_offset)
,所有 4 个角,这就是我们有这么多循环的原因。在外部两个
i
(行)、j
(列)循环中,我们迭代整个网格,该(i, j)
位置将是矩形top-left
的点1s
,我们需要迭代整个网格,因为所有可能的1s
矩形可能位于整个网格的top-left
任何点。(row, col)
在循环开始时,j
我们检查位置处的网格应(i, j)
始终包含1
,因为在循环内部我们仅搜索所有矩形1s
。k
循环遍历矩形的所有可能top-right
位置。如果等于,我们应该跳出循环,因为没有必要进一步向右扩展,因为这样的矩形将始终包含。(i, k)
1s
(i, k)
0
k
0
在之前的循环中,我们固定了
top-left
矩形top-right
的角点。现在我们需要搜索两个底角。为此,我们需要以不同角度向下延伸矩形,直到到达第一个0
。off
循环尝试以所有可能的角度向下延伸矩形(0
(垂直垂直),+1
(45
从上到下向右移动的度数),-1
(-45
度)),off
基本上是grid[y][x]
“上方”的数字(对应于 byY
)grid[y + 1][x + off]
。l
Y
尝试以不同角度向下(方向)延伸矩形off
。它被扩展到第一个,0
因为它不能进一步扩展(因为每个这样的矩形已经包含0
)。循环内部
l
有一个if grid[l][max(0, j + off * (l - i)) : min(k + 1 + off * (l - i), c)] != ones[:k - j + 1]:
条件,基本上这if
是为了检查矩形的最后一行是否包含所有内容,1
如果不是,if
则会跳出循环。此条件比较两个list
切片是否相等。矩形的最后一行从点(l, j + angle_offset)
(表达式max(0, j + off * (l - i))
,最大限制为0 <= X
)到点(l, k + angle_offset)
(表达式min(k + 1 + off * (l - i), c)
,最小限制为X < c
)。在循环内部
l
还有其他行,ry, rx = l, k + off * (l - i)
计算bottom-right
矩形的点,(ry, rx)
即(l, k + angle_offset)
,该(ry, rx)
位置用于存储在ma
数组中找到的最大值,该数组存储所有找到的最大矩形,包含有关在点ma[ry][rx]
处的矩形的信息。bottom-right
(ry, rx)
rv = (l + 1 - i, k + 1 - j, off)
line 计算ma[ry][rx]
数组条目的新的可能候选者,之所以可能,ma[ry][rx]
是因为仅当新候选者具有更大的面积时才更新1s
。这里元组rv[0]
内的值rv
包含height
这样的矩形,rv[1]
包含width
这样的矩形(width
等于矩形底行的长度),rv[2]
包含这样的矩形的角度。条件
if rv[0] * rv[1] > ma[ry][rx][0] * ma[ry][rx][1]:
及其主体仅检查rv
面积是否大于数组内的当前最大值ma[ry][rx]
,如果大于则更新此数组条目(ma[ry][rx] = rv
)。我会提醒您,其中包含有关当前找到的最大面积矩形的ma[ry][rx]
信息,该矩形具有、和。(width, height, angle)
bottom-right
(ry, rx)
width
height
angle
完毕!算法运行后,数组
ma
包含有关所有最大面积有角度的矩形(簇)的信息1s
,以便所有簇都可以恢复并稍后打印到控制台。Largest of all such1s
-clusters 等于 somerv0 = ma[ry0][rx0]
,只需迭代一次 的所有元素ma
并找到这样的点(ry0, rx0)
,以使ma[ry0][rx0][0] * ma[ry0][rx0][1]
(面积)最大。然后最大的簇将有bottom-right
point(ry0, rx0)
、bottom-left
point(ry0, rx0 - rv0[1] + 1)
、top-right
point(ry0 - rv0[0] + 1, rx0 - rv0[2] * (rv0[0] - 1))
、top-left
point(ry0 - rv0[0] + 1, rx0 - rv0[1] + 1 - rv0[2] * (rv0[0] - 1))
(这里只是角度偏移,即第一行与矩形的最后一行相比rv0[2] * (rv0[0] - 1)
移动了多少)。X
# ----------------- Main function solving task -----------------
def solve(
grid, *,
algo = 1, # Choose algorithm, 0 - Simple, 1 - Advanced, 2 - Simple-ListComp
check = True, # If True run all algorithms and check that they produce same results, otherwise run just chosen algorithm without checking
text = False, # If true then grid is a multi-line text (string) having grid elements separated by spaces
print_ = True, # Print results to console
show_non_max = True, # When printing if to show all clusters, not just largest, as B, C, D, E... (chars from "cchars")
cchars = ['.'] + [chr(ii) for ii in range(ord('B'), ord('Z') + 1)], # Clusters-chars, these chars are used to show clusters from largest to smallest
one = None, # Value of "one" inside grid array, e.g. if you have grid with chars then one may be equal to "1" string. Defaults to 1 (for non-text) or "1" (for text).
offs = [0, +1, -1], # All offsets (angles) that need to be checked, "off" is such that grid[i + 1][j + off] corresponds to next row of grid[i][j]
debug = False, # If True, extra debug info is printed
):
# Preparing
assert algo in [0, 1, 2], algo
if text:
grid = [l.strip().split() for l in grid.splitlines() if l.strip()]
if one is None:
one = 1 if not text else '1'
r, c = len(grid), len(grid[0])
sgrid = '\n'.join([''.join([str(grid[ii][jj]) for jj in range(c)]) for ii in range(r)])
mas, ones = [], [one] * max(c, r)
# ----------------- Simple Algorithm, O(N^5) Complexity -----------------
if algo == 0 or check:
ma = [[(0, 0, 0) for jj in range(c)] for ii in range(r)] # Array containing maximal answers, Lower-Right corners
for i in range(r):
for j in range(c):
if grid[i][j] != one:
continue
for k in range(j + 1, c): # Ensure at least 2 ones along X
if grid[i][k] != one:
break
for off in offs:
for l in range(i + 1, r): # Ensure at least 2 ones along Y
if grid[l][max(0, j + off * (l - i)) : min(k + 1 + off * (l - i), c)] != ones[:k - j + 1]:
l -= 1
break
ry, rx = l, k + off * (l - i)
rv = (l + 1 - i, k + 1 - j, off)
if rv[0] * rv[1] > ma[ry][rx][0] * ma[ry][rx][1]:
ma[ry][rx] = rv
mas.append(ma)
ma = None
# ----------------- Advanced Algorithm using Dynamic Programming, O(N^3) Complexity -----------------
if algo == 1 or check:
ma = [[(0, 0, 0) for jj in range(c)] for ii in range(r)] # Array containing maximal answers, Lower-Right corners
for off in offs:
d = [[(0, 0, 0) for jj in range(c)] for ii in range(c)]
for i in range(r):
f, d_ = 0, [[(0, 0, 0) for jj in range(c)] for ii in range(c)]
for j in range(c):
if grid[i][j] != one:
f = j + 1
continue
if f >= j:
# Check that we have at least 2 ones along X
continue
df = [(0, 0, 0) for ii in range(c)]
for k in range(j, -1, -1):
t0 = d[j - off][max(0, k - off)] if 0 <= j - off < c and k - off < c else (0, 0, 0)
if k >= f:
t1 = (t0[0] + 1, t0[1], off) if t0 != (0, 0, 0) else (0, 0, 0)
t2 = (1, j - k + 1, off)
t0 = t1 if t1[0] * t1[1] >= t2[0] * t2[1] else t2
# Ensure that we have at least 2 ones along Y
t3 = t1 if t1[0] > 1 else (0, 0, 0)
if k < j and t3[0] * t3[1] < df[k + 1][0] * df[k + 1][1]:
t3 = df[k + 1]
df[k] = t3
else:
t0 = d_[j][k + 1]
if k < j and t0[0] * t0[1] < d_[j][k + 1][0] * d_[j][k + 1][1]:
t0 = d_[j][k + 1]
d_[j][k] = t0
if ma[i][j][0] * ma[i][j][1] < df[f][0] * df[f][1]:
ma[i][j] = df[f]
d = d_
mas.append(ma)
ma = None
# ----------------- Simple-ListComp Algorithm using List Comprehension, O(N^5) Complexity -----------------
if algo == 2 or check:
ma = [
[
max([(0, 0, 0)] + [
(h, w, off)
for h in range(2, i + 2)
for w in range(2, j + 2)
for off in offs
if all(
cr[
max(0, j + 1 - w - off * (h - 1 - icr)) :
max(0, j + 1 - off * (h - 1 - icr))
] == ones[:w]
for icr, cr in enumerate(grid[max(0, i + 1 - h) : i + 1])
)
], key = lambda e: e[0] * e[1])
for j in range(c)
]
for i in range(r)
]
mas.append(ma)
ma = None
# ----------------- Checking Correctness and Printing Results -----------------
if check:
# Check that we have same answers for all algorithms
masx = [[[cma[ii][jj][0] * cma[ii][jj][1] for jj in range(c)] for ii in range(r)] for cma in mas]
assert all([masx[0] == e for e in masx[1:]]), 'Maximums of algorithms differ!\n\n' + sgrid + '\n\n' + (
'\n\n'.join(['\n'.join([' '.join([str(e1).rjust(2) for e1 in e0]) for e0 in cma]) for cma in masx])
)
ma = mas[0 if not check else algo]
if print_:
cchars = ['.'] + [chr(ii) for ii in range(ord('B'), ord('Z') + 1)] # These chars are used to show clusters from largest to smallest
res = [[grid[ii][jj] for jj in range(c)] for ii in range(r)]
mac = [[ma[ii][jj] for jj in range(c)] for ii in range(r)]
processed = set()
sid = 0
for it in range(r * c):
sma = sorted(
[(mac[ii][jj] or (0, 0, 0)) + (ii, jj) for ii in range(r) for jj in range(c) if (ii, jj) not in processed],
key = lambda e: e[0] * e[1], reverse = True
)
if len(sma) == 0 or sma[0][0] * sma[0][1] <= 0:
break
maxv = sma[0]
if it == 0:
maxvf = maxv
processed.add((maxv[3], maxv[4]))
show = True
for trial in [True, False]:
for i in range(maxv[3] - maxv[0] + 1, maxv[3] + 1):
for j in range(maxv[4] - maxv[1] + 1 - (maxv[3] - i) * maxv[2], maxv[4] + 1 - (maxv[3] - i) * maxv[2]):
if trial:
if mac[i][j] is None:
show = False
break
elif show:
res[i][j] = cchars[sid]
mac[i][j] = None
if show:
sid += 1
if not show_non_max and it == 0:
break
res = '\n'.join([''.join([str(res[ii][jj]) for jj in range(c)]) for ii in range(r)])
print(
'Max:\nArea: ', maxvf[0] * maxvf[1], '\nSize Row,Col: ', (maxvf[0], maxvf[1]),
'\nLowerRight Row,Col: ', (maxvf[3], maxvf[4]), '\nAngle: ', ("-1", " 0", "+1")[maxvf[2] + 1], '\n', sep = ''
)
print(res)
if debug:
# Print all computed maximums, for debug purposes
for cma in [ma, mac]:
print('\n' + '\n'.join([' '.join([f'({e0[0]}, {e0[1]}, {("-1", " 0", "+1")[e0[2] + 1]})' for e0_ in e for e0 in (e0_ or ('-', '-', 0),)]) for e in cma]))
print(end = '-' * 28 + '\n')
return ma
# ----------------- Testing -----------------
def test():
# Iterating over text inputs or other ways of producing inputs
for s in [
"""
1 1 0 0 0 1 0 1
1 1 1 0 1 1 1 1
1 0 0 0 1 0 1 1
0 0 1 0 1 0 1 1
1 1 1 1 0 0 1 1
0 0 1 1 1 1 1 0
0 1 0 0 1 0 1 1
""",
"""
1 0 1 1 0 1 0 0
0 1 1 0 1 0 0 1
1 1 0 0 0 0 0 1
0 1 1 1 0 1 0 1
0 1 1 1 1 0 1 1
1 1 0 0 0 1 0 0
0 1 1 1 0 1 0 1
""",
"""
0 1 1 0 1 0 1 1
0 0 1 1 0 0 0 1
0 0 0 1 1 0 1 0
1 1 0 0 1 1 1 0
0 1 1 0 0 1 1 0
0 0 1 0 1 0 1 1
1 0 0 1 0 0 0 0
0 1 1 0 1 1 0 0
"""
]:
solve(s, text = True)
if __name__ == '__main__':
test()
输出:
Max:
Area: 8
Size Row,Col: (4, 2)
LowerRight Row,Col: (4, 7)
Angle: 0
CC000101
CC1011..
100010..
001010..
1BBB00..
00BBBDD0
010010DD
----------------------------
Max:
Area: 6
Size Row,Col: (3, 2)
LowerRight Row,Col: (2, 1)
Angle: -1
10..0100
0..01001
..000001
0BBB0101
0BBB1011
CC000100
0CC10101
----------------------------
Max:
Area: 12
Size Row,Col: (6, 2)
LowerRight Row,Col: (5, 7)
Angle: +1
0..01011
00..0001
000..010
BB00..10
0BB00..0
001010..
10010000
01101100
----------------------------
添加回答
举报