查找二维直方图的峰 [英] Find peak of 2d histogram

问题描述

我对某些(x, y)数据进行了二维直方图处理，并得到了如下图所示的图像:

I make a 2d histogram of some (x, y) data and I get an image like this one:

我想要一种方法来获取将最大值存储在H中的点的(x, y)坐标.例如，在上图的情况下，它将是两个具有近似坐标的点:(1090, 1040)和(1110, 1090).

I want a way to get the (x, y) coordinates of the point(s) that store the maximum values in H. For example, in the case of the image above it would be two points with the aprox coordinates: (1090, 1040) and (1110, 1090).

这是我的代码:

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.cm as cm
from os import getcwd
from os.path import join, realpath, dirname

# Path to dir where this code exists.
mypath = realpath(join(getcwd(), dirname(__file__)))
myfile = 'datafile.dat'

x, y = np.loadtxt(join(mypath,myfile), usecols=(1, 2), unpack=True)

fig = plt.figure()
ax = fig.add_subplot(111)

xmin, xmax = min(x), max(x)
ymin, ymax = min(y), max(y)

rang = [[xmin, xmax], [ymin, ymax]]

binsxy = [int((xmax - xmin) / 20), int((ymax - ymin) / 20)]

H, xedges, yedges = np.histogram2d(x, y, range=rang, bins=binsxy)

extent = [yedges[0], yedges[-1], xedges[0], xedges[-1]]
cp = ax.imshow(H.transpose()[::-1], interpolation='nearest', extent=extent, cmap=cm.jet)
fig.colorbar(cp)

plt.show()

---

修改

我已经尝试了Marek和qarma发布的解决方案，试图获取垃圾箱的坐标而不是它们的索引，就像这样:

I've tried the solutions posted by Marek and qarma attempting to obtain the coordinates of the bins rather than the index of them, like so:

# Marek's answer
x_cent, y_cent = unravel_index(H.argmax(), H.shape)
print('Marek')
print(x_cent, y_cent)
print(xedges[x_cent], yedges[y_cent])

# qarma's answer
idx = list(H.flatten()).index(H.max())
x_cent2, y_cent2 = idx / H.shape[1], idx % H.shape[1]
local_maxs = np.argwhere(H == H.max())
print('\nqarma')
print(x_cent2, y_cent2)
print(xedges[x_cent2], yedges[y_cent2])
print(xedges[local_maxs[0,0]], yedges[local_maxs[0,1]], xedges[local_maxs[1,0]], yedges[local_maxs[1,1]])

其结果是:

Marek
(53, 50)
(1072.7838144329899, 1005.0837113402063)

qarma
(53, 50)
(1072.7838144329899, 1005.0837113402063)
(1072.7838144329899, 1005.0837113402063, 1092.8257731958763, 1065.3611340206187)

所以最大坐标是相同的，这很好！现在，我遇到了一个小问题，因为如果放大2d图，我会发现全局最大值和局部最大值的坐标都稍微偏离中心:

So the maximum coordinates are the same which is good! Now I have a small issue because if I zoom in on the 2d plot, I see that the coordinates are a little off-centered for both the global maximum and the local maximum:

这是为什么?

推荐答案

在这里，您可以找到第一个全局最大值

Here's how you can find first global maximum

idx = list(H.flatten()).index(H.max())
x, y = idx / H.shape[1], idx % H.shape[1]

剩下的所有最大值的查找坐标留给读者练习...

Finding coordinate of all maxima was left as exercise to the reader...

numpy.argwhere(H == H.max())

修改

您的代码:

H, xedges, yedges = np.histogram2d(x, y, range=rang, bins=binsxy)

此处H包含直方图值和直方图bin的xedges, yedges边界.请注意，edges数组的大小比相应维度中H的大小大一.因此:

Here H contains histogram values and xedges, yedges boundaries for histogram bins. Note that size of edges arrays is one larger than size of H in corresponding dimension. Thus:

for x, y in numpy.argwhere(H == H.max()):
    # center is between x and x+1
    print numpy.average(xedges[x:x + 2]), numpy.average(yedges[y:y + 2])

这篇关于查找二维直方图的峰的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！