[SciPy-User] multidimensional least squares fitting

josef.pktd@gmai... josef.pktd@gmai...
Wed Nov 9 13:12:54 CST 2011


On Wed, Nov 9, 2011 at 1:39 PM, Daniel Mader
<danielstefanmader@googlemail.com> wrote:
> Hi Josef,
>
> thanks a lot for this quick answer, I'll definitely have a look at
> your suggestion!
>
> However, I'd be very interested if this is possible with leastsq, too?
> Maybe I don't need it right now, but you never you what's next in life
> :)

there is also optimize.curve_fit as wrapper for least_sq

your error function returns 2d instead of 1d

def __residual(params, f,  x, y):
   '''
   Define fit function;
   Return residual error.
   '''
   p0, p1, p2 = params
   #x,y,f = map(np.ravel, (x,y,f))
   return np.ravel(p0 + p1*x + p2*y - f)

you need one ravel
I didn't look at the rest

Josef


>
> Thanks again,
> greetings,
>
> Daniel
>
>
>
> 2011/11/9  <josef.pktd@gmail.com>:
>> On Wed, Nov 9, 2011 at 12:12 PM, Daniel Mader
>> <danielstefanmader@googlemail.com> wrote:
>>> Hi everyone,
>>>
>>> I'd like to do some rather simple multidimensional curve fitting.
>>> Simple, because usually it's only a plane, or a weak 2nd order
>>> surface.
>>>
>>> Here's the same question which I tried to follow, but I have no clue
>>> how to feed the 2D arrays into leastsq():
>>> http://stackoverflow.com/questions/529184/simple-multidimensional-curve-fitting
>>
>> If you just want to estimate a multiple linear or polynomial function,
>> then optimize.leastsq is overkill, linalg is enough.
>>
>> The premade solution is to use:
>>
>> statsmodels.sourceforge.net/generated/scikits.statsmodels.regression.linear_model.OLS.html
>>
>> Josef
>>
>>>
>>> Likely I am just missing a small piece of information, and I'd be
>>> happy to get a clue :)
>>>
>>> Thanks in advance,
>>> and here's some code to demonstrate what I want and to get started,
>>> Daniel
>>>
>>>
>>> import pylab
>>> import scipy
>>> import scipy.optimize
>>>
>>> from mpl_toolkits.mplot3d import Axes3D
>>>
>>> #import sys,os,platform
>>> #if platform.system() == 'Windows':
>>> #    home = os.environ['HOMESHARE']
>>> #elif platform.system() == 'Linux':
>>> #    home = os.environ['HOME']
>>> #sys.path.append(home + '/python')
>>> #sys.path.append(home + '/11_PythonWork')
>>> #import pylabSettings
>>>
>>> ##******************************************************************************
>>> ##******************************************************************************
>>>
>>> '''
>>> f = p0 + p1*x + p2*y
>>> '''
>>>
>>> ##------------------------------------------------------------------------------
>>> def __residual(params, f, x, y):
>>>    '''
>>>    Define fit function;
>>>    Return residual error.
>>>    '''
>>>    p0, p1, p2 = params
>>>    return p0 + p1*x + p2*y - f
>>>
>>> ## load raw data (=create some dummy data):
>>> dataX = scipy.arange(0,11,1)
>>> dataY = dataX/10.
>>> dataZ = 0.5 + 1.1*dataX + 1.5*dataY
>>> dataXX, dataYY = scipy.meshgrid(dataX,dataY)
>>> dataZZ = 0.5 + 1.1*dataXX + 1.5*dataYY
>>>
>>> ## plot data
>>> pylab.close('all')
>>> fig = pylab.figure()
>>> ax = Axes3D(fig)
>>> ax.plot_wireframe(dataXX, dataYY, dataZZ)
>>> ax.set_xlabel('x')
>>> ax.set_ylabel('y')
>>> ax.set_zlabel('z')
>>> pylab.show()
>>>
>>> ## guess initial values for parameters
>>> p0 = [0., 1., 1.]
>>> print __residual(p0, dataZZ, dataXX, dataYY)
>>>
>>> ## works but is not 2D!
>>> p1, p_cov = scipy.optimize.leastsq(__residual, x0=p0, args=(dataZ,
>>> dataX, dataY))
>>> print p1
>>>
>>> ## doesn't work :()
>>> p1, p_cov = scipy.optimize.leastsq(__residual, x0=p0, args=(dataZZ,
>>> dataXX, dataYY))
>>> print p1
>>> _______________________________________________
>>> SciPy-User mailing list
>>> SciPy-User@scipy.org
>>> http://mail.scipy.org/mailman/listinfo/scipy-user
>>>
>> _______________________________________________
>> SciPy-User mailing list
>> SciPy-User@scipy.org
>> http://mail.scipy.org/mailman/listinfo/scipy-user
> _______________________________________________
> SciPy-User mailing list
> SciPy-User@scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>


More information about the SciPy-User mailing list