0

我编写了一个 python 脚本，它打开 UV-Vis 光谱并尝试用大量函数来拟合它们。但是，我希望在找到最小残差时将拟合步骤显示在图中。Stackoverflow 实际上有一些涉及这个想法的示例（http://stackoverflow.com/questions/4098131/matplotlib-update-a-plot），但由于某种原因，这种方法对我来说效果不佳。我所说的“工作不太好”的意思是绘图窗口不响应脚本中发生的更新。我试图将我的代码缩减为更易于理解、仍然可以编译的代码，但也比示例更接近我的代码，如下所示。

重新表述我的问题：有没有更好的方法通过拟合过程来刷新这种类型的屏幕，以使窗口不会变成“（不响应）”？

这是我的简化代码：

# import modules that I'm using
import matplotlib
matplotlib.use('GTKAgg')
import Tkinter
from Tkinter import *
import numpy as np
import scipy as sc
import matplotlib.pyplot as pltlib
# lmfit is imported becuase parameters are allowed to depend on each other along with bounds, etc.
from lmfit import minimize, Parameters, Minimizer

#If button is pressed on the window, open a file and get the data
def open_File():
    # file is opened here and some data is taken
    # I've just set some arrays here so it will compile alone
    x=[]
    y=[]
    for num in range(0,1000):x.append(num*.001+1)
    # just some random function is given here, the real data is a UV-Vis spectrum
    for num2 in range(0,1000):y.append(sc.math.sin(num2*.06)+sc.math.e**(num2*.001))
    X = np.array(x)
    Y = np.array(y)

    # plot the initial data in one figure
    pltlib.ion()
    pltlib.interactive(True)
    pltlib.figure(1)
    pltlib.plot(X,Y, "r-")
    pltlib.show()

    #deconvolute this initial data into deveral lorentzian profiles
    deconvolute(X,Y)

#lorentz line for constructing the more complex function
def lorentz(x, amp, center, width):
    return amp*1/sc.math.pi*(width/((x-center)**2+width**2))

def deconvolute(X,Y):
    #make 2nd figure for the refreshing screen
    deconvFig = pltlib.figure(2)
    ax = deconvFig.add_subplot(111)
    line1,line2=ax.plot(X,Y,'r-',X,Y,'r-')

    # setup parameters for several (here is 30, I potentially hae many more in the real program)
    params = Parameters()
    for p in range(0,30):
        params.add('amp' + str(p), value=1)
        params.add('center' + str(p), value=1)
        params.add('width' + str(p), value=1)

    #get residual function for minimizing
    def residual(params, X, data=None):
        model = 0
        # get values for each lorentz and sum them up
        for p in range(0,30):
            amp = params['amp' + str(p)].value
            center = params['center' + str(p)].value
            width = params['width' + str(p)].value
            tmpLorentz = lorentz(X, amp, center, width)
            model = model + tmpLorentz

        # This is where the main problem is.
        # This 2nd figure stops responding after a very small (1?) number of iterations
        ########################################
        # I want redraw the figure at every step through the fitting process
        line2.set_ydata(model)
        deconvFig.canvas.draw()
        print 'screen should be refreshed'
        ########################################

        return (data - model)

    #fit the function to the data
    result = minimize(residual, params, args=(X, Y))
    print 'done fitting the program'

#create a window with a button
MainWindow = Tk()
Button(text='Open a File', command=open_File).pack(side=BOTTOM)
MainWindow.mainloop()
4

1 回答 1

0

有趣的是，我尝试运行一个简单的测试。

import time
from matplotlib import pyplot as pltlib
deconvFig = pltlib.figure(2)
ax = deconvFig.add_subplot(111)
X, Y = range(10), range(10)
line1,line2 = ax.plot(X,Y,'r-',X,Y,'r-')
for x in xrange(2, 6, 1):
    line2.set_ydata(range(0, 10*x, x))
    deconvFig.canvas.draw()
    time.sleep(2)

>>> import matplotlib
>>> matplotlib.__version__
'1.1.0'

并且它按预期工作。
也许是因为您生成了第二个数字。

import time
from matplotlib import pyplot as pltlib

pltlib.ion()
pltlib.interactive(True)
pltlib.figure(1)
pltlib.plot(range(10),range(10), "r-")
pltlib.show()

deconvFig = pltlib.figure(2)
ax = deconvFig.add_subplot(111)
X, Y = range(10), range(10)
line1,line2 = ax.plot(X,Y,'r-',X,Y,'r-')
for x in xrange(2, 6, 1):
    line2.set_ydata(range(0, 10*x, x))
    deconvFig.canvas.draw()
    time.sleep(2)

不，仍然可以正常工作。
这可能是我的设置。

虽然它也有可能以非常慢的速度最小化，所以当你绘制更新时你无法分辨出差异，你可以计算 RMSE 看看差异有多大

print numpy.sqrt(numpy.sum((data - model)**2)/model.shape[0])/numpy.mean(data) * 100

另外我通常使用 scipy 的最小化功能http://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize.html因为它可以最小化大多数功能，尽管它通过随机改变输入来工作，所以我不知道它能有多快，但它可以应用在很多很多情况下。

我希望这有帮助。

于 2012-06-28T07:25:48.710 回答