Numpy二维标量场网格过滤结果异常的原因探究

我正在尝试理解为何标注为"naive filtering"的代码块无法提取二维标量场的预期角落区域，而标注为"correct result"的代码块可以实现该效果。相关代码如下：

import numpy as np
import matplotlib.pyplot as plt

# creating the underlying x and y axes
nx, ny = 101, 111
x = np.linspace(-1, 1, nx)
y = np.linspace(-1, 1, ny)
X, Y = np.meshgrid(x, y, indexing='xy')

# computing the scalar 2D field
f = np.exp(-1.1*X*X - 0.9*Y*Y + 0.7*X*Y)

# saving figure
fig, ax = plt.subplots()
ax.imshow(f)
plt.savefig('field_full.png')

# data filters
mx = x > 0
my = y < 0
MX, MY = np.meshgrid(mx, my, indexing='xy')

# naive filtering
mf = (MX & MY)
lf = f[mf].reshape(len(x[mx]), len(y[my]))

# saving figure
fig, ax = plt.subplots()
ax.imshow(lf)
plt.savefig('field_filtered_naive.png')

# correct result
mf = (MX & MY).T
lf = f.T[mf].reshape(len(x[mx]), len(y[my]))

# saving figure
fig, ax = plt.subplots()
ax.imshow(lf)
plt.savefig('field_filtered_correct.png')

编辑：在看到@Mercury的回答后，我明确了需求：我清楚如何得到预期结果，但困惑的是"naive filtering"代码块为何无法直接生效。此外，@Mercury提出的另一种可行代码：

import numpy as np
import matplotlib.pyplot as plt

# creating the underlying x and y axes
nx, ny = 101, 111
x = np.linspace(-1, 1, nx)
y = np.linspace(-1, 1, ny)
X, Y = np.meshgrid(x, y, indexing='xy')

# computing the scalar 2D field
f = np.exp(-1.1*X*X - 0.9*Y*Y + 0.7*X*Y)

# saving figure
fig, ax = plt.subplots()
ax.imshow(f)
plt.savefig('field_full.png')

# data filters
mx = x > 0
my = y < 0
MX, MY = np.meshgrid(mx, my, indexing='xy')

# filtered field data
mf = (MX & MY)
reduced_nx = len(x[mx])
reduced_ny = len(y[my])
lf = f[mf].reshape(reduced_ny, reduced_nx)

# saving figure
fig, ax = plt.subplots()
ax.imshow(lf)
plt.savefig('field_filtered.png')

这让我更困惑，我特意定义了reduced_nx和reduced_ny来凸显问题：原场f的维度为nx×ny，但过滤后的lf维度却需设为reduced_ny×reduced_nx而非预期的相反顺序，这种"隐性转置"看似设计缺陷，易导致难以察觉的错误。

内容的提问来源于stack exchange，提问作者ragdollfun