Getting off-diagonal elements in dataframe

By following the pandas DataFrame Chart , I can get the diagonal elements using np.diag. How can I get off-diagonal elements in a data frame (assuming the data frame size is nxn)

+4
source share
3 answers

I will use the same @Matt data file xf

xf = pd.DataFrame(np.random.rand(5, 5))

However, I will indicate that if the diagonal is zero, the use np.diag(np.diag(xf)) != 0will be violated.

A way to ensure that you are masking the diagonal is to evaluate if the row indices are not equal for the column indices.

Option 1
numpy.indices

, numpy np.indices.

, .

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]]

... .

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]]

, ,

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

pd.DataFrame(
    np.where(np.equal(*np.indices(xf.shape)), np.nan, xf.values),
    xf.index, xf.columns
)

2
numpy.arange slice

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

%%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
+5

, np.eye, :

xf = pd.DataFrame(np.random.rand(5,5))
xf.mask(np.eye(5, dtype = bool))
+4

EDITED , SomeGuy-, , 0.

xf = pd.DataFrame(np.random.rand(5,5))
xf.mask(np.eye(5, dtype = bool))

, 0

True False, /.

xf = pd.DataFrame(np.random.rand(5,5))
diag = np.diag(np.diag(xf))
xf.mask(diag != 0)
0

Source: https://habr.com/ru/post/1680533/


All Articles