在数据框中获取非对角元素

use*_*827 5 python numpy pandas

pandas DataFrame对角线之后,我可以使用对角线元素np.diag.如何获取数据帧中的非对角元素(假设数据帧大小为nxn)

Som*_*Guy 6

使用生成的掩码,np.eye如:

xf = pd.DataFrame(np.random.rand(5,5))
xf.mask(np.eye(5, dtype = bool))
Run Code Online (Sandbox Code Playgroud)


piR*_*red 5

我将使用@Matt的相同数据帧 xf

xf = pd.DataFrame(np.random.rand(5, 5))
Run Code Online (Sandbox Code Playgroud)

但是,我会指出,如果对角线恰好等于零,则使用np.diag(np.diag(xf)) != 0将会中断.

保证屏蔽对角线的方法是评估行索引是否不等于列索引.

选项1
numpy.indices

方便地,numpy通过np.indices功能提供这些.

观察它们的样子

rows, cols = np.indices((5, 5))

print(rows)

[[0 0 0 0 0]
 [1 1 1 1 1]
 [2 2 2 2 2]
 [3 3 3 3 3]
 [4 4 4 4 4]]

print(cols)

[[0 1 2 3 4]
 [0 1 2 3 4]
 [0 1 2 3 4]
 [0 1 2 3 4]
 [0 1 2 3 4]]
Run Code Online (Sandbox Code Playgroud)

它们相等的地方......对角线.

print((cols == rows).astype(int))

[[1 0 0 0 0]
 [0 1 0 0 0]
 [0 0 1 0 0]
 [0 0 0 1 0]
 [0 0 0 0 1]]
Run Code Online (Sandbox Code Playgroud)

因此,通过这些,我们可以掩盖它们相等的位置

xf.mask(np.equal(*np.indices(xf.shape)))

          0         1         2         3         4
0       NaN  0.605436  0.573386  0.978588  0.160986
1  0.295911       NaN  0.509203  0.692233  0.717464
2  0.275767  0.966976       NaN  0.883339  0.143704
3  0.628941  0.668836  0.468928       NaN  0.309901
4  0.286933  0.523243  0.693754  0.253426       NaN
Run Code Online (Sandbox Code Playgroud)

我们可以使用更快一点

pd.DataFrame(
    np.where(np.equal(*np.indices(xf.shape)), np.nan, xf.values),
    xf.index, xf.columns
)
Run Code Online (Sandbox Code Playgroud)

选项2
numpy.arange与切片分配

v = xf.values.copy()
i = j = np.arange(np.min(v.shape))
v[i, j] = np.nan
pd.DataFrame(v, xf.index, xf.columns)

          0         1         2         3         4
0       NaN  0.605436  0.573386  0.978588  0.160986
1  0.295911       NaN  0.509203  0.692233  0.717464
2  0.275767  0.966976       NaN  0.883339  0.143704
3  0.628941  0.668836  0.468928       NaN  0.309901
4  0.286933  0.523243  0.693754  0.253426       NaN
Run Code Online (Sandbox Code Playgroud)
%%timeit 
v = xf.values.copy()
i = j = np.arange(np.min(v.shape))
v[i, j] = np.nan
pd.DataFrame(v, xf.index, xf.columns)


%timeit pd.DataFrame(np.where(np.eye(np.min(xf.shape)), np.nan, xf.values), xf.index, xf.columns)
%timeit pd.DataFrame(np.where(np.equal(*np.indices(xf.shape)), np.nan, xf.values), xf.index, xf.columns)
%timeit xf.mask(np.equal(*np.indices(xf.shape)))
%timeit xf.mask(np.diag(np.diag(xf.values)) != 0)
%timeit xf.mask(np.eye(np.min(xf.shape), dtype=bool)

10000 loops, best of 3: 74.5 µs per loop
10000 loops, best of 3: 85.7 µs per loop
10000 loops, best of 3: 77 µs per loop
1000 loops, best of 3: 519 µs per loop
1000 loops, best of 3: 517 µs per loop
1000 loops, best of 3: 528 µs per loop
Run Code Online (Sandbox Code Playgroud)