假设您想将每行的所有元素设置为零,直到最后一个负元素(根据问题中列出的样例输出),这里可以提供两种方法。
方法一
这种方法基于
np.cumsum
生成要设置为零的元素掩码,如下所示 -
# Get boolean mask with TRUEs for each row starting at the first element and
# ending at the last negative element
mask = (np.cumsum(A2[:,::-1]<0,1)>0)[:,::-1]
# Use mask to set all such al TRUEs to zeros as per the expected output in OP
A2[mask] = 0
样例运行 -
In [280]: A2 = np.random.randint(-4,10,(6,7)) # Random input 2D array
In [281]: A2
Out[281]:
array([[-2, 9, 8, -3, 2, 0, 5],
[-1, 9, 5, 1, -3, -3, -2],
[ 3, -3, 3, 5, 5, 2, 9],
[ 4, 6, -1, 6, 1, 2, 2],
[ 4, 4, 6, -3, 7, -3, -3],
[ 0, 2, -2, -3, 9, 4, 3]])
In [282]: A2[(np.cumsum(A2[:,::-1]<0,1)>0)[:,::-1]] = 0 # Use mask to set zeros
In [283]: A2
Out[283]:
array([[0, 0, 0, 0, 2, 0, 5],
[0, 0, 0, 0, 0, 0, 0],
[0, 0, 3, 5, 5, 2, 9],
[0, 0, 0, 6, 1, 2, 2],
[0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 9, 4, 3]])
方法二
这个方法起始于从@tom10的答案
中找到最后一个负元素的索引,并发展成使用broadcasting
的掩码查找方法,以获得与方法一
类似的所需输出。
last_idx = A2.shape[1] - 1 - np.argmax(A2[:,::-1]<0, axis=1)
invalid_idx = A2[np.arange(A2.shape[0]),last_idx]>=0
last_idx[invalid_idx] = -1
mask = np.arange(A2.shape[1]) < (last_idx[:,None] + 1)
A2[mask] = 0
运行时测试 -
函数定义:
def broadcasting_based(A2):
last_idx = A2.shape[1] - 1 - np.argmax(A2[:,::-1]<0, axis=1)
last_idx[A2[np.arange(A2.shape[0]),last_idx]>=0] = -1
A2[np.arange(A2.shape[1]) < (last_idx[:,None] + 1)] = 0
return A2
def cumsum_based(A2):
A2[(np.cumsum(A2[:,::-1]<0,1)>0)[:,::-1]] = 0
return A2
运行时:
In [379]: A2 = np.random.randint(-4,10,(100000,100))
...: A2c = A2.copy()
...:
In [380]: %timeit broadcasting_based(A2)
10 loops, best of 3: 106 ms per loop
In [381]: %timeit cumsum_based(A2c)
1 loops, best of 3: 167 ms per loop
验证结果 -
In [384]: A2 = np.random.randint(-4,10,(100000,100))
...: A2c = A2.copy()
...:
In [385]: np.array_equal(broadcasting_based(A2),cumsum_based(A2c))
Out[385]: True
[-3,-4,-5,3,4,-7,8]
必须给我[0,0,0,0,0,0,8]
。 - Thomas Leonard