Home
pNumeric module for Python is set of functions and datatypes for signal processing focused on Kalman filter utilization.
The Kalman filter is created from scratch and all support routines are also provided. So you do not need any additional libraries - Numeric, etc. There is also FFT algorithm available.
Because it deals with matrix algebra, some support algorithms had to be implemented. There are functions for matrix manipulation and matrix operations - matrix multiplication, determinant computing, matrix inversion, etc.
No 3rd party libraries are needed while using pNumeric.
Early versions has been tested even for PyS60 (Python for Symbian S60). There are build scripts still available.
pNumeric uses distutils, so the installation on PC is pretty simple - just run:
python setup.py install
Note that you'll need root privilegies to install it. You can just build it by typing
python setup.py install
and copy pnumeric.so to the application's path.
There are also build scripts for PyS60 available.
Note that POSIX standard operating system and tools are essential. Build environment must be installed. Preferred is Linux operating system with GNU GCC.
Module pNumeric has several methods. The table gives a brief desctiption:
- ones(n) - returns Matrix of size n with all items set to 1
- zeros(n) - returns Matrix of size n with all items set to 0
- eye(n) - returns eye Matrix of size n with zeros everywhere except ones at main diagonal
- kf_process(...) - Kalman filter function
- fft(Vector v) - Fast Fourier Transform of v
- mean(Vector v) - mean value of v
- rms(Vector v) - Root Mean Square of v
Vector datatype is used to store measurements or other vectors. To create Vector object just simply type:
>>> from pnumeric import *
>>> v = Vector([1, 2, 3])
>>> v
[1.0, 2.0, 3.0]
Vector constructor accepts list or tuple with all numeric values. It is possible to get Vertor item by accessing it as list:
>>> v[2]
3.0
Also changing any of items is possible:
>>> v[1] = 88
>>> v
[1.0, 88.0, 3.0]
Matrix is used to store matrixes and implements matrix algebra related functions.
Use following syntax to create Matrix object:
from pnumeric import *
m2 = Matrix([ [.1, .2], [-.2, .1] ])
m3 = Matrix([ [.2, .4, .2], [-.2, .2, .0], [.2, .2, -.2] ])
Vector constructor accepts list or tuple with all numeric values.
We can use also tuple instead of lists - be sure you specify the same length of sublists (subtuples).
Matrix object has following methods:
- det() - returns determinant of the matrix
- inv() - returns inversion matrix
Matrix object has following properties:
- rows - returns number of Matrix columns
- cols - returns number of Matrix columns
- shape - returns an Matrix shape tuple (rows, cols)
It is possible to get Matrix item by accessing as list. This returns matrix as a Vector type:
>>> m3[2]
[-0.2, 0.2, 0.0]
Also change any of items is possible:
>>> m3[1][1] = 66
>>> m3
0.2 0.4 0.2
-0.2 66 0
0.2 This function parameters are step number an model parameters.
Then we can simply put the function callback name as the last parameter to kf_process
and filter will automaticaly call the function after every step.0.2 -0.2
Kalman filter in pNumeric package is designed as general as possible. We can filter MIMO systems with a nuber of state variables. Also updating of model after every time step is possible (EKF).
Following example is to show Kalman filter usage. This is the simplest case of estimating a constant of 501.45 disturbed by white noise.
from pnumeric import *
from random import gauss
length = 30 # data length
value = 501.45 # constant value to be estimated
t = range(length)
# generate some noise around the value
yv = [gauss(value, 9) for x in range(length)]
y = Matrix([yv])
u = Matrix([ [0]*length ]) # input values are not necessary - so create Matrix of zeros
# define system matrixes
A = Matrix([ [1] ])
B = Matrix([ [0] ])
C = Matrix([ [1] ])
D = Matrix([ [0] ])
# process and noise variations
Q = Matrix([ [0.5] ])
R = Matrix([ [1.0] ])
# define initial values
x0 = Matrix([ [500.0] ])
P0 = Matrix([ [9.0] ])
# process input by Kalman filter
x_est, y_est, P_est = kf_process(A, B, C, D, y, u, x0, P0, Q, R)
x_est now contains list of estimated states, y_est is list of estimated output values and P_est is list of covariance estimations.
To update the model, define a Python function:
# callback for updating the matrixes
def callback(step, A, B, C, D, x):
A[0][1] = step*dt
B[0][0] = 0.5*(step*dt)**2
B[1][0] = step*dt
# reset the x
x[0][0] = s0
x[1][0] = v0
This function parameters are step number an model parameters. Then we can simply put the function callback name as the last parameter to kf_process and filter will automaticaly call the function after every step.
fft() function computes a spectrum of time series given by a Vector parameter.
Example:
v = Vector([0, 1, 0, -1, 0, 1, 0, -1]) # create time series Vector
out = fft(v)
fft() returns a Vector containing sperctrum of time series.
pNumeric module also contains some statistical functions:
- mean(Vector v) - returns mean value of v
- rms(Vector v) - returns Root Mean Square of v