python多维数组拟合_类似二维数组:shape为(12,9)，python中有哪些函数可以进行二维高斯拟合吗？...

我的理解是，你的二维矩阵是个 histogram. 里面都是 int 对吧？

这样的话，方法也许不唯一。但我提供一个从 histogram 里 recover data 的思路。为了方便理解，我先模拟一遍你观测到的数据，然后再拟合。以下代码是数据生产的模拟过程。

import matplotlib.pyplot as plt

import numpy as np

from scipy.stats import multivariate_normal

# simulate the data with some fake mean and cov

mean = np.array([1., 3.])

cov = np.array([[3., 1.],

[1., 3.]])

# this is the true MVN we want to fit from data

mvn = multivariate_normal(mean=mean, cov=cov)

# the observed data

data = mvn.rvs(size=200)

# the histogram `hist` is your (9, 12) matrix

hist, xe, ye = np.histogram2d(rvs[:,0], rvs[:,1], bins=[9, 12])

# show the histogram and true density

xx, yy = np.meshgrid(xe, ye, indexing='ij')

pos = np.dstack([xx, yy])

density = mvn.pdf(pos)

_, (ax1, ax2) = plt.subplots(ncols=2)

ax1.imshow(hist)

ax2.imshow(density)

ax1.set_title("histogram")

ax2.set_title("true density")

好，现在假设我得到的这个 histogram 就是你的那个矩阵。我该怎么你和右边的 true density 呢？我这里用的方式是先从 histogram 里还原出那200个二维随机矢量，然后对 MVN 做极大似然估计就行了。看下面这个函数就行。

def hist_to_data_array(hist, xedges, yedges):

"""

recover data from histogram.

Parameters

----------

hist: 2D histogram, shape (9, 12)

xedges: bin edges in x direction

yedges: bin edges in y direction

Returns

-------

data: shape (N, 2), where n = hist.sum()

"""

n = int(hist.sum())

data_array = np.empty(shape=(n, 2), dtype=np.float)

nx = len(xedges) - 1

ny = len(yedges) - 1

k = 0

for i in range(nx):

for j in range(ny):

m = int(hist[i,j])

data_array[k:k+m] = [(xedges[i] + xedges[i+1])/2., (yedges[j] + yedges[j+1])/2.]

k += m

return data_array

有了这个函数之后，剩下的就简单了。

# notice the hist, xe and ye are from np.histogram2d in the first code block

data_recovered = hist_to_data_array(hist, xe, ye)

# estimate the mean vector and cov matrix

mean_fit = np.mean(data_recovered , axis=0)

cov_fit = np.cov(data_recovered , rowvar=0)

# show the result

mvn_fit = multivariate_normal(mean=mean_fit, cov=cov_fit)

density_fit = mvn_fit.pdf(pos)

plt.imshow(density_fit)

plt.title("fitted density")

这样基本就OK了。注意我在模拟数据的过程中，histogram 这一步是有 xe, ye 的。 如果你手上的数据只是一个 histogram, 没有这些 bin edge 的信息，那么在调用函数的时候把xe设为range(0, 10), ye设置成range(0, 13)就行了。就像下面这样。注意10=9+1, 13=12+1因为它们是edge，比bins多1. 当然之后你的坐标系也要相应调整。

xe = np.arange(0, 10)

ye = np.arange(0, 13)

data_recovered = hist_to_data_array(hist, xe , ye)