numpy4

最新推荐文章于 2023-11-23 13:04:25 发布

难受想吃面条

最新推荐文章于 2023-11-23 13:04:25 发布

阅读量260

点赞数

分类专栏： python学习笔记文章标签： numpy

原文链接：http://datawhale.club/t/topic/186

版权

python学习笔记专栏收录该内容

6 篇文章 0 订阅

订阅专栏

numpy线性代数

import numpy as np

numpy.dot(a, b, out)

a: array_like
b: array_like

# a, b都是一维数组，结果为内积
x = np.array([i for i in range(1, 6)])
y = np.array([i for i in range(2, 7)])
z = np.dot(x, y)
print(z)

x = np.array([[1, 2, 3], [3, 4, 5], [6, 7, 8]])
y = np.array([[5, 4, 2], [1, 7, 9], [0, 4, 5]])
z = np.dot(x, y)
print(z)

[[  7  30  35]
 [ 19  60  67]
 [ 37 105 115]]

z =  np.dot(y, x)
print(z)

[[ 29  40  51]
 [ 76  93 110]
 [ 42  51  60]]

矩阵特征值与特征向量

numpy.linalg.eig(a)

a: 要计算的方阵
return: w: (…,M)array 特征值，无序
v: (…,M,M)array 归一化的特征向量（单位长度）
numpy.linalg.eigvals(a)
计算特征值

# 创建一个对角矩阵
x = np.diag((1,2,3))
print(x)
print(np.linalg.eigvals(x))

[[1 0 0]
 [0 2 0]
 [0 0 3]]
[1. 2. 3.]

a,b = np.linalg.eig(x)
print(a)
print("*****"*4)
print(b)

[1. 2. 3.]
********************
[[1. 0. 0.]
 [0. 1. 0.]
 [0. 0. 1.]]

# numpy.allclose(a, b, rtol=1e-05, atol=1e-08, equal_nan=False)
# 如果两个array在误差范围内按元素相等，返回True
for i in range(3):
    if np.allclose(a[i]*b[:, i], np.dot(x, b[:, i])):
        print("Right")
    else:
        print("Error")

Right
Right
Right

判断对称阵是否为正定阵(特征值是否全部为正)

A = np.arange(16).reshape(4,4)
print(A)

[[ 0  1  2  3]
 [ 4  5  6  7]
 [ 8  9 10 11]
 [12 13 14 15]]

A = A+A.T  # 将方阵转换成对称阵，矩阵加矩阵的转置
print(A)

[[ 0  5 10 15]
 [ 5 10 15 20]
 [10 15 20 25]
 [15 20 25 30]]

B = np.linalg.eigvals(A)
print(B)

[ 6.74165739e+01 -7.41657387e+00  1.82694656e-15 -1.72637110e-15]

if np.all(B>0):
    print("Yes")
else:
    print("No")

No

矩阵分解

奇异值分解

u, s, v = numpy.linalg.svd(a, full_matrices=True,compute_uv=True,hermitial=False)
a: (…, M, N) array_like一个实数或复数矩阵， a.ndim>=2
full_matrices: bool, optional 如果为True，u的形状(…, M,M),vh的形状为(…, N,N),否则形状为(…, M,K) and (…, N,K) k=min(M,N)
hermitian: bool, optional 如果为True，则假定a为Hermitian（如果为实值，则为对称），从而启用一种更有效的方法来查找奇异值。默认为False。

return:
u: (…, M,M), (…,M,K) array
s: (…, K) array 矩阵a的奇异值分解，除了对角元素，均为0，对角元素降序排列
vh: (…,N,N), (…,N,K) array

A = np.array([[4, 11, 14], [8, 7, -2]])
print(A)

[[ 4 11 14]
 [ 8  7 -2]]

u, s, vh = np.linalg.svd(A, full_matrices=False)
print(u.shape)
print(u)

(2, 2)
[[ 0.9486833  -0.31622777]
 [ 0.31622777  0.9486833 ]]

print(s.shape)
print(s)

(2,)
[18.97366596  9.48683298]

print(vh.shape)
print(vh)

(2, 3)
[[ 0.33333333  0.66666667  0.66666667]
 [ 0.66666667  0.33333333 -0.66666667]]

a = np.dot(u, np.diag(s))
print(a)
a = np.dot(a, vh)
print(a)

[[18. -3.]
 [ 6.  9.]]
[[ 4. 11. 14.]
 [ 8.  7. -2.]]

A = np.array([[1, 1], [1, -2], [2, 1]])
print(A)

[[ 1  1]
 [ 1 -2]
 [ 2  1]]

u,s,vh = np.linalg.svd(A, full_matrices=False)
print(u.shape)
print(u)

(3, 2)
[[-5.34522484e-01 -1.11022302e-16]
 [ 2.67261242e-01 -9.48683298e-01]
 [-8.01783726e-01 -3.16227766e-01]]

print(s.shape)
print(s)

(2,)
[2.64575131 2.23606798]

print(vh.shape)
print(vh)

(2, 2)
[[-0.70710678 -0.70710678]
 [-0.70710678  0.70710678]]

a = np.dot(u, np.diag(s))
a = np.dot(a, vh)
print(a)

[[ 1.  1.]
 [ 1. -2.]
 [ 2.  1.]]

QR分解

numpy.linalg.qr(a, mode=‘reduced’)
mode: {‘reduced’, ‘complete’, ‘r’, ‘raw’}, optional
mode = reduced：返回(M, N)的列向量两两正交的矩阵q，和(N, N)的三角阵r（Reduced QR分解）。
mode = complete：返回(M, M)的正交矩阵q，和(M, N)的三角阵r（Full QR分解）。

A = np.array([[2, -2, 3], [1, 1, 1], [1, 3, -1]])
print(A)

[[ 2 -2  3]
 [ 1  1  1]
 [ 1  3 -1]]

q,r = np.linalg.qr(A)
print(q.shape)
print(q)

(3, 3)
[[-0.81649658  0.53452248  0.21821789]
 [-0.40824829 -0.26726124 -0.87287156]
 [-0.40824829 -0.80178373  0.43643578]]

print(r.shape)
print(r)

(3, 3)
[[-2.44948974  0.         -2.44948974]
 [ 0.         -3.74165739  2.13808994]
 [ 0.          0.         -0.65465367]]

print(np.dot(q,r))

[[ 2. -2.  3.]
 [ 1.  1.  1.]
 [ 1.  3. -1.]]

a = np.allclose(np.dot(q,q.T), np.eye(3))
print(a)

True

A = np.array([[1, 1], [1, -2], [2, 1]])
print(A)

[[ 1  1]
 [ 1 -2]
 [ 2  1]]

q,r = np.linalg.qr(A,mode='complete')
print(q.shape)
print(q)

(3, 3)
[[-0.40824829  0.34503278 -0.84515425]
 [-0.40824829 -0.89708523 -0.16903085]
 [-0.81649658  0.27602622  0.50709255]]

print(r.shape)
print(r)

(3, 2)
[[-2.44948974 -0.40824829]
 [ 0.          2.41522946]
 [ 0.          0.        ]]

print(np.dot(q,r))

[[ 1.  1.]
 [ 1. -2.]
 [ 2.  1.]]

q,r = np.linalg.qr(A)
print(q.shape)
print(q)

(3, 2)
[[-0.40824829  0.34503278]
 [-0.40824829 -0.89708523]
 [-0.81649658  0.27602622]]

print(np.allclose(np.dot(q.T, q), np.eye(2)))
# q为正交阵

True

Cholesky分解

L= numpy.linalg.cholesky(a)
返回正定矩阵a的Cholesky分解 a = L*L.T,L是下三角阵

A = np.array([[1, 1, 1, 1], [1, 3, 3, 3],
              [1, 3, 5, 5], [1, 3, 5, 7]])
print(A)

[[1 1 1 1]
 [1 3 3 3]
 [1 3 5 5]
 [1 3 5 7]]

np.linalg.eigvals(A)

array([13.13707118,  1.6199144 ,  0.51978306,  0.72323135])

L = np.linalg.cholesky(A)
print(L)

[[1.         0.         0.         0.        ]
 [1.         1.41421356 0.         0.        ]
 [1.         1.41421356 1.41421356 0.        ]
 [1.         1.41421356 1.41421356 1.41421356]]

print(np.dot(L, L.T))

[[1. 1. 1. 1.]
 [1. 3. 3. 3.]
 [1. 3. 5. 5.]
 [1. 3. 5. 7.]]

范数和其它矩阵特征

numpy.linalg.norm(x, ord=None, axis=None,keepdim=False)
ord $\quad\quad\quad$ 矩阵范数 $\quad\quad\quad$ 向量范数
None $\quad\quad\quad$ Frobenius norm$\quad\quad\quad $2 - n o r m^{'} f r o^{'}$ \quad\quad\quad $F r o b e n i u s n o r m$ \quad\quad\quad $-^{'} n u c^{'}$ \quad\quad\quad $n u c l e a r n o r m$ \quad\quad\quad $- i n f$ \quad\quad\quad $m a x (s u m (a b s (x), a x i s = 1))$ \quad\quad\quad $m a x (a b s (x)) - i n f$ \quad\quad\quad $m i n (s u m (a b s (x), a x i s = 1))$ \quad\quad\quad $m i n (a b s (x)) 0$ \quad\quad\quad $-$ \quad\quad $s u m (x! = 0) 1$ \quad\quad\quad $m a x (s u m (a b s (x), a x i s = 0))$ \quad\quad\quad $a s b e l o w - 1$ \quad\quad\quad $m i n (s u m (a b s (x), a x i s = 0))$ \quad\quad\quad $a s b e l o w 2$ \quad\quad\quad $2 - n o r m (l a r g e s t s i n g . v a l u e)$ \quad\quad\quad $a s b e l o w - 2$ \quad\quad\quad $s m a l l e s t s i n g u l a r v a l u e$ \quad\quad\quad $a s b e l o w o t h e r$ \quad\quad\quad $-$ \quad\quad\quad\quad$sum(abs(x)ord)(1./ord)

x = np.array([1, 2, 3, 4])
print(np.linalg.norm(x, ord=1))

10.0

print(np.sum(np.abs(x)))

print(np.linalg.norm(x, ord=2))
print(np.sum(np.abs(x)**2)**0.5)

5.477225575051661
5.477225575051661

print(np.linalg.norm(x, ord=-np.inf))
print(np.min(np.abs(x)))

1.0
1

print(np.linalg.norm(x, ord=np.inf)) 
print(np.max(np.abs(x)))

4.0
4

# 矩阵范数
A = np.array([[1, 2, 3, 4], [2, 3, 5, 8],
              [1, 3, 5, 7], [3, 4, 7, 11]])
print(A)

[[ 1  2  3  4]
 [ 2  3  5  8]
 [ 1  3  5  7]
 [ 3  4  7 11]]

# 每列之和的最大值
print(np.linalg.norm(A, ord=1))  
print(np.max(np.sum(A, axis=0)))

30.0
30

print(np.linalg.norm(A, ord=2)) 
print(np.max(np.linalg.svd(A, compute_uv=False)))

20.24345358700576
20.24345358700576

# 每行之和的最大值
print(np.linalg.norm(A, ord=np.inf)) 
print(np.max(np.sum(A, axis=1)))

25.0
25

print(np.linalg.norm(A, ord='fro')) 
print(np.sqrt(np.trace(np.dot(A.T, A))))

20.273134932713294
20.273134932713294

方阵的行列式

numpy.det(a)

x = np.array([[1, 2], [3, 4]])

print(np.linalg.det(x))

-2.0000000000000004

矩阵的秩

numpy.linalg.matrix_rank(M, tol=None, hermitian=False) 返回矩阵的秩。

I = np.eye(3)
print(I)

[[1. 0. 0.]
 [0. 1. 0.]
 [0. 0. 1.]]

r = np.linalg.matrix_rank(I)
print(r)

矩阵的迹

numpy.trace(a, offset=0, axis1=0, axis2=1, dtype=None, out=None) 方阵的迹就是主对角元素之和。

x = np.array([[1, 2, 3], [3, 4, 5], [6, 7, 8]])
print(x)

[[1 2 3]
 [3 4 5]
 [6 7 8]]

print(np.trace(x))

解方程和逆矩阵

numpy.linalg.inv(a) 计算矩阵a的逆矩阵（矩阵可逆的充要条件：det(a) != 0，或者a满秩）。

A = np.array([[1, -2, 1], [0, 2, -1], [1, 1, -2]])
print(A)

[[ 1 -2  1]
 [ 0  2 -1]
 [ 1  1 -2]]

A_inverse = np.linalg.inv(A)
print(A_inverse)

[[ 1.00000000e+00  1.00000000e+00 -1.11022302e-16]
 [ 3.33333333e-01  1.00000000e+00 -3.33333333e-01]
 [ 6.66666667e-01  1.00000000e+00 -6.66666667e-01]]

x = np.allclose(np.dot(A,A_inverse), np.eye(3))
print(x)

True

A_det = np.linalg.det(A) #行列式
# 求伴随矩阵
A_companion = A_inverse*A_det
print(A_companion)

[[-3.00000000e+00 -3.00000000e+00  3.33066907e-16]
 [-1.00000000e+00 -3.00000000e+00  1.00000000e+00]
 [-2.00000000e+00 -3.00000000e+00  2.00000000e+00]]

求解线性方程组

numpy.linalg.solve(a, b) 求解线性方程组或矩阵方程。

#  x + 2y +  z = 7
# 2x -  y + 3z = 7
# 3x +  y + 2z =18

A = np.array([[1, 2, 1], [2, -1, 3], [3, 1, 2]])
b = np.array([7, 7, 18])
x = np.linalg.solve(A, b)
print(x)

[ 7.  1. -2.]

y = np.allclose(np.dot(A,x),b)
print(y)

True

# 计算两个数组的欧式距离
a = np.array([1, 2, 3, 4, 5])
b = np.array([4, 5, 6, 7, 8])
dis = np.sum((b-a)**2)**0.5
print(dis)

6.708203932499369

A = [[4,-1,1],[-1,3,-2],[1,-2,3]]
a,b = np.linalg.eig(A)
print(a)
print("***"*3)
print(b)

[3. 6. 1.]
*********
[[-8.16496581e-01 -5.77350269e-01  3.89855447e-17]
 [-4.08248290e-01  5.77350269e-01  7.07106781e-01]
 [ 4.08248290e-01 -5.77350269e-01  7.07106781e-01]]

难受想吃面条

关注

0
点赞
踩
0

收藏

觉得还不错? 一键收藏
0
评论
numpy4

numpy线性代数import numpy as npnumpy.dot(a, b, out)a: array_likeb: array_like# a, b都是一维数组，结果为内积x = np.array([i for i in range(1, 6)])y = np.array([i for i in range(2, 7)])z = np.dot(x, y)print(z)70x = np.array([[1, 2, 3], [3, 4, 5], [6, 7, 8]])y
复制链接

扫一扫