Question

我有一个kalman过滤器的实现，我正在进行矩阵运算。在某些时候，我应该减去两个1x1矩阵。我有一个错误，我不知道它来自哪里。在python中进行矩阵运算的最佳方法是什么？

import numpy as np
import pylab as pl
import scipy as Sci
import scipy.linalg as linalg

class GetPos(object):
    def __init__(self):
        self.Posp = 0
        self.Velp = 80
        self.z = np.matrix(0)
    def __repr__(self):
        return "from GetPos.__repr__ z=%s" % (self.z)
    def __call__(self):
        self.dt = 0.1
        self.w = 0 + 10*np.random.random()
        self.v = 0 + 10*np.random.random()            
        self.z = self.Posp + self.Velp*self.dt + self.v
        self.Posp = self.z - self.v
        self.Velp = 80 + self.w
        print 'from GetPos.__call__ z = %s' % self.z
        return self.z

class DvKalman(object):
    def __init__(self):
        self.dt = .1
        self.A = np.matrix([[1., self.dt],[0,1]])
        self.H = np.matrix([1., 0])
        self.Q = np.matrix([[1,0.],[0,3]])
        self.R = np.matrix(10)
        self.x = np.matrix([0,20]).T
        self.P = np.matrix(5*np.eye(2))
        #print 'P matrix \n%s' % self.P
        self.firstRun = 0
    def __call__(self, z):
        self.z = z
        print 'from DvKalman.__call__ slef.z = %s and z = %s' % (self.z,z)
        self.xp = self.A * self.x
        self.Pp = self.A*self.P*self.A.T  + self.Q
        self.K = self.Pp * self.H.T * linalg.inv(np.absolute(self.H*self.Pp*self.H.T + self.R));
        print 'from DvKalman.__call__  z=%s, \npreviouse x=\n%s \nH = \n%s \nand P=\n%s \nand xp=\n%s,\n Pp = \n%s,\n K=\n%s' % (self.z,self.x,self.H, self.P,self.xp,self.Pp,self.K)
        newM1 = self.H*self.xp    
        print 'This is self.H*self.xp %s and this is self.z = %s' % (newM1, self.z)
        newM2 = self.z - self.H*self.xp
        print 'This should give simple substruction %s' % newM2                 
        self.x = self.xp + self.K*(self.z - self.H*self.xp)
        self.P = self.Pp - self.K*self.H*self.Pp
        print 'new values x=%s and P=%s' % (self.x,self.P)
        return (self.x)
def TestDvKalman():
    Nsamples = np.arange(0,10,.1)

    kal = DvKalman()
    #print type(kal)
    Xsaved = []
    Zsaved = []

    for i in range(len(Nsamples)):
        z = GetPos()
        print z
        print 'from TestDvKalman zpos = %s' % z
        Zsaved.append(z)
        [position, velocity] = kal(z)
        print position, velocity
        Xsaved.append([position, velocity])
    print Zsaved
    print Xsaved
#    f1 = pl.subplot(121)
#    f1 = pl.plot(Xsaved, 'x-',label = 'Xsaved')
#    f1 = pl.legend()
#    
#    f2 = pl.subplot(122)
#    f2 = pl.title('Kalman Velocity')
#    f2 = pl.plot(Zsaved, 'o-', color = 'brown',label = 'Zsaved')
#    f2 = pl.legend()
#    
#    pl.show()

if __name__ == '__main__':  
    TestDvKalman()

我添加了一些print行并调试代码，我添加了新的变量newM，这些不在代码中。矩阵打印正确This is self.H*self.xp [[ 2.]] and this is self.z = from GetPos.__repr__ z=[[0]]两个矩阵都是1x1，但我仍然收到错误，不知道为什么。错误是：

    newM2 = self.z - self.H*self.xp
TypeError: unsupported operand type(s) for -: 'GetPos' and 'matrix'

我怀疑我在某处乱搞类型，但不知道在哪里以及如何纠正它。你能指出我错误的位置以及如何构建这样的代码以避免将来出现类似的错误吗？

Answer 1

您正在将GetPos实例传递给DvKalman __call__方法。所以你试图减去一个GetPos实例和一个矩阵。不是矩阵和矩阵。

Answer 2

在TestDvKalman中，此行

    z = GetPos()

将z设置为GetPos的实例。您可以在此行中将其用作kal的参数：

    [position, velocity] = kal(z)

因此__call__ DvKalman方法的GetPos方法被赋予了self.z的实例，您将其另存为newM2 = self.z - self.H*self.xp。这导致了这一行的错误：

{{1}}

Answer 3

将newM2 = self.z - self.H*self.xp替换为newM2 = self.z() - self.H*self.xp。

程序应该可以使用（但我无法确认它是否真的想要你想要的）

python / numpy中的矩阵减法

3 个答案: