如何将那些保持在 sns.heatmap 的对角线上？

Question

I want to plot the correlation Matrix with sns.heatmap and have some questions. This is my code:

plt.figure(figsize=(8,8)) mask =np.zeros_like(data.corr()) mask[np.triu_indices_from(mask)] = True sns.heatmap(data.corr(), mask=mask, linewidth=1, annot=True, fmt=".2f",cmap='coolwarm',vmin=-1, vmax=1) plt.show()

and this is what i get: [Correlation Matrix][1] [1]: https://i.stack.imgur.com/DX2oN.png \

Now i have some questions:

1) How can i keep the ones in the diagonale?

2) How can i change the position of the x-axis?

3) I want that the colorbar goes from 1 till -1, but the code is not working

I hope someone can help.

Thx

Answer 1

I think you have to check data.corr() , because your code is correct and gives the diagnoal (see below). One question is: you use np.triu but the picture you show displays np.tirl .

Here the code I've tested - the diagonal is there:

N = 5
A = np.arange(N*N).reshape(N,N)

B = np.tril(A)

mask =np.zeros_like(A)
mask[np.triu_indices_from(mask)] = True

print('A'); print(A); print()
print('tril(A)'); print(B); print()
print('mask'); print(mask); print()

gives

A
[[ 0  1  2  3  4]
 [ 5  6  7  8  9]
 [10 11 12 13 14]
 [15 16 17 18 19]
 [20 21 22 23 24]]

tril(A)
[[ 0  0  0  0  0]
 [ 5  6  0  0  0]
 [10 11 12  0  0]
 [15 16 17 18  0]
 [20 21 22 23 24]]

mask
[[1 1 1 1 1]
 [0 1 1 1 1]
 [0 0 1 1 1]
 [0 0 0 1 1]
 [0 0 0 0 1]]

edit: suplement

you could re-fine the mask, eg

C = A *mask
D = np.where(C > 1, 1,C)
print('D'); print(D)

gives

D
[[0 1 1 1 1]
 [0 1 1 1 1]
 [0 0 1 1 1]
 [0 0 0 1 1]
 [0 0 0 0 1]]

The first element of the diagonal of D is now a Zero since the first element of the diagonal of A is a Zero too.

edit: suplement 2

F = np.tril(A,-1)
E = np.eye(N)
G = E + F

print('F'); print(F); print()
print('E'); print(E); print()
print('G'); print(G); print()

gives

F
[[ 0  0  0  0  0]
 [ 5  0  0  0  0]
 [10 11  0  0  0]
 [15 16 17  0  0]
 [20 21 22 23  0]]

E
[[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.]]

G
[[ 1.  0.  0.  0.  0.]
 [ 5.  1.  0.  0.  0.]
 [10. 11.  1.  0.  0.]
 [15. 16. 17.  1.  0.]
 [20. 21. 22. 23.  1.]]

Answer 2

mask[np.triu_indices_from(mask)] will define the triangle (including diagonal)

mask[np.eye(mask.shape[0], dtype=bool)] will define the diagonal.

If you put those together, you can control them independently. (Be aware you need to set the triangle before the diagonal).

def plot_correlation_matrix(df, remove_diagonal=True, remove_triangle=False, **kwargs):
    corr = df.corr()
    # Apply mask
    mask = np.zeros_like(corr, dtype=np.bool)
    mask[np.triu_indices_from(mask)] = remove_triangle
    mask[np.eye(mask.shape[0], dtype=bool)] = remove_diagonal
    # Plot
    # plt.figure(figsize=(8,8))
    sns.heatmap(corr, mask=mask, **kwargs)
    plt.show()

So this command will generate the matrix, removing the upper triangle, but keeping the diagonal:

plot_correlation_matrix(df[colunas_notas], remove_diagonal=False, remove_triangle=True)

Answer 3

Change of the position of the x-axis

Since I'm not experienced with seaborn I would use matplotlib to plot the heat map ( here an example ) an then use matplotlib's twinx() or twiny() to place the axis where you want to have it ( here an example ).

(I think that can be done with seaborn too - I just do not know it)