Geek Answers Handbook

Python 2D Gaussian Fit with NaN values ​​in data

I am very new to Python, but I am trying to create a 2D Gaussian fit for some data. In particular, stellar fluxes are associated with certain positions in the coordinate system / grid. However, not all positions in my grid have corresponding flow values. I don’t want these values ​​to be zero if it suits me, but I can’t set them to nanand still get a Gaussian approach to work. This is the code I'm using (slightly modified from here ):

import numpy
import scipy
from numpy import *
from scipy import optimize

def gaussian(height, center_x, center_y, width_x, width_y):
    width_x = float(width_x)
    width_y = float(width_y)
    return lambda x,y: height*exp(-(((center_x-x)/width_x)**2+((center_y-y)/width_y)**2)/2)

def moments(data):
    total = nansum(data)
    X, Y = indices(data.shape)
    center_x = nansum(X*data)/total
    center_y = nansum(Y*data)/total
    row = data[int(center_x), :]
    col = data[:, int(center_y)]
    width_x = nansum(sqrt(abs((arange(col.size)-center_y)**2*col))/nansum(col))
    width_y = nansum(sqrt(abs((arange(row.size)-center_x)**2*row))/nansum(row))
    height = nanmax(data)
    return height, center_x, center_y, width_x, width_y

def fitgaussian(data):
    params = moments(data)
    errorfunction = lambda p: ravel(gaussian(*p)(*indices(data.shape)) - data)
    p, success = optimize.leastsq(errorfunction, params)
    return p

parameters = fitgaussian(data)
fit = gaussian(*parameters)

2D- data. , 0 nan , parameters [nan nan nan nan nan]. , ! , . !

+4
2

- NaNs data. 2D X, Y:

X, Y = np.indices(data.shape)
mask = ~np.isnan(data)
x = X[mask]
y = Y[mask]
data = data[mask]

optimize.leastsq ( , optimize.curve_fit) :

p, success = optimize.leastsq(errorfunction, params, args=(x, y, data))

, data NaNs

data = make_data(shape)


import matplotlib.pyplot as plt
plt.imshow(data)
plt.show()

enter image description here

, , NaN,

import numpy as np
from scipy import optimize
np.set_printoptions(precision=4)


def gaussian(p, x, y):
    height, center_x, center_y, width_x, width_y = p
    return height*np.exp(-(((center_x-x)/width_x)**2+((center_y-y)/width_y)**2)/2)

def moments(data):
    total = np.nansum(data)
    X, Y = np.indices(data.shape)
    center_x = np.nansum(X*data)/total
    center_y = np.nansum(Y*data)/total
    row = data[int(center_x), :]
    col = data[:, int(center_y)]
    width_x = np.nansum(np.sqrt(abs((np.arange(col.size)-center_y)**2*col))
                        /np.nansum(col))
    width_y = np.nansum(np.sqrt(abs((np.arange(row.size)-center_x)**2*row))
                        /np.nansum(row))
    height = np.nanmax(data)
    return height, center_x, center_y, width_x, width_y

def errorfunction(p, x, y, data):
    return gaussian(p, x, y) - data

def fitgaussian(data):
    params = moments(data)
    X, Y = np.indices(data.shape)
    mask = ~np.isnan(data)
    x = X[mask]
    y = Y[mask]
    data = data[mask]
    p, success = optimize.leastsq(errorfunction, params, args=(x, y, data))
    return p

def make_data(shape):
    h, w = shape
    p = 50, h/2.0, w/2.0, h/3.0, w/5.0
    print('Actual parameters: {}'.format(np.array(p)))
    X, Y = np.indices(shape)
    data = gaussian(p, X, Y) + np.random.random(shape)
    mask = np.random.random(shape) < 0.3
    data[mask] = np.nan
    return data

shape = 100, 200
data = make_data(shape)
X, Y = np.indices(shape)
parameters = fitgaussian(data)
print('Fitted parameters: {}'.format(parameters))
fit = gaussian(parameters, X, Y)


Actual parameters: [  50.       50.      100.       33.3333   40.    ]
Fitted parameters: [ 50.2908  49.9992  99.9927  33.7039  40.6149]
+3

, . , y.

, , ''

# assumes data.shape = (1, 3) where data[:,0:1] is the x,y axis
# data[:,2] contains the flux values
data = numpy.delete(data, numpy.where(data[:,3] == ''), axis=0)

, nan

data = numpy.delete(data, numpy.where(data[:,3] == numpy.nan), axis=0)
+1

More articles:


All Articles