算法导论4.2Strassen矩阵算法

Strassen的思想是把矩阵分块，用分块矩阵的和来生成七个中间矩阵后，再做乘法，因此复杂度为n的lg7次方。注意它在计算非整数时，会有累积误差。目前最优的是Coppersmith和Winograd提出的算法，复杂度是n的2.376次方。

python代码

import numpy as np
def StrassenSquareMatrixMultiple(A,B):
    n = A.shape[0]
    if n == 1:
        return np.dot(A,B)
    S1 = B[0:n//2,n//2:n] - B[n//2:n,n//2:n]
    S2 = A[0:n//2,0:n//2] + A[0:n//2,n//2:n]
    S3 = A[n//2:n,0:n//2] + A[n//2:n,n//2:n]
    S4 = B[n//2:n,0:n//2] - B[0:n//2,0:n//2]
    S5 = A[0:n//2,0:n//2] + A[n//2:n,n//2:n]
    S6 = B[0:n//2,0:n//2] + B[n//2:n,n//2:n]
    S7 = A[0:n//2,n//2:n] - A[n//2:n,n//2:n]
    S8 = B[n//2:n,0:n//2] + B[n//2:n,n//2:n]
    S9 = A[0:n//2,0:n//2] - A[n//2:n,0:n//2]
    S10 = B[0:n//2,0:n//2] + B[0:n//2,n//2:n]

    P1 = StrassenSquareMatrixMultiple(A[0:n//2,0:n//2],S1)
    P2 = StrassenSquareMatrixMultiple(S2,B[n//2:n,n//2:n])
    P3 = StrassenSquareMatrixMultiple(S3,B[0:n//2,0:n//2])
    P4 = StrassenSquareMatrixMultiple(A[n//2:n,n//2:n],S4)
    P5 = StrassenSquareMatrixMultiple(S5,S6)
    P6 = StrassenSquareMatrixMultiple(S7,S8)
    P7 = StrassenSquareMatrixMultiple(S9,S10)
    C = np.zeros([n,n])
    C[0:n//2,0:n//2] = P5+P4-P2+P6
    C[0:n//2,n//2:n] = P1 + P2
    C[n//2:n,0:n//2] = P3+P4
    C[n//2:n,n//2:n] = P5+P1-P3-P7
    return C
A = np.array([[1,2],
              [3,4]])
B = np.array([[5,6],
              [7,8]])
print(StrassenSquareMatrixMultiple(A,B))