使用 pandas.shift() 基于 scipy.signal.correlate 对齐数据集
声明：本页面是StackOverFlow热门问题的中英对照翻译，遵循CC BY-SA 4.0协议，如果您需要使用它，必须同样遵循CC BY-SA许可，注明原文地址和作者信息，同时你必须将它归于原作者（不是我）：StackOverFlow
原文地址: http://stackoverflow.com/questions/19642443/
Warning: these are provided under cc-by-sa 4.0 license. You are free to use/share it, But you must attribute it to the original authors (not me):
StackOverFlow
Use of pandas.shift() to align datasets based on scipy.signal.correlate
提问by not link
I have datasets that look like the following: data0, data1, data2(analogous to time versus voltage data)
我有如下所示的数据集：data0、data1、data2（类似于时间与电压数据）
If I load and plot the datasets using code like:
如果我使用如下代码加载和绘制数据集：
import pandas as pd
import numpy as np
from scipy import signal
from matplotlib import pylab as plt
data0 = pd.read_csv('data0.csv')
data1 = pd.read_csv('data1.csv')
data2 = pd.read_csv('data2.csv')
plt.plot(data0.x, data0.y, data1.x, data1.y, data2.x, data2.y)
I get something like:
我得到类似的东西：
now I try to correlate data0 with data1:
现在我尝试将 data0 与 data1 相关联：
shft01 = np.argmax(signal.correlate(data0.y, data1.y)) - len(data1.y)
print shft01
plt.figure()
plt.plot(data0.x, data0.y,
data1.x.shift(-shft01), data1.y)
fig = plt.gcf()
with output:
带输出：
-99
and
和
which works just as expected! but if I try it the same thing with data2, I get a plot that looks like:
正如预期的那样工作！但是如果我用 data2 尝试同样的事情，我会得到一个看起来像的图：
with a positive shift of 410. I think I am just not understanding how pd.shift()works, but I was hoping that I could use pd.shift()to align my data sets. As far as I understand, the return from correlate()tells me how far off my data sets are, so I should be able to use shift to overlap them.
的正偏移410。我想我只是不明白是如何pd.shift()工作的，但我希望我可以pd.shift()用来对齐我的数据集。据我了解， return fromcorrelate()告诉我我的数据集有多远，所以我应该能够使用 shift 来重叠它们。
回答by HYRY
panda.shift()is not the correct method to shift curve along x-axis. You should adjust X values of the points:
panda.shift()不是沿 x 轴移动曲线的正确方法。您应该调整点的 X 值：
plt.plot(data0.x, data0.y)
for target in [data1, data2]:
dx = np.mean(np.diff(data0.x.values))
shift = (np.argmax(signal.correlate(data0.y, target.y)) - len(target.y)) * dx
plt.plot(target.x + shift, target.y)
here is the output:
这是输出：
回答by AhabTheArab
@HYRYone correction to your answer: there is an indexing mismatch between len(), which is one-based, and np.argmax(), which is zero-based. The line should read:
@HYRY对您的答案进行一次更正：len()基于 1 的 和基于np.argmax()零的之间存在索引不匹配。该行应为：
shift = (np.argmax(signal.correlate(data0.y, target.y)) - (len(target.y)-1)) * dx
shift = (np.argmax(signal.correlate(data0.y, target.y)) - (len(target.y)-1)) * dx
For example, in the case where your signals are already aligned:
例如，在您的信号已经对齐的情况下：
len(target.y)= N (one-based)
len(target.y)= N（从一开始）
The cross-correlation function has length 2N-1, so the center value, for aligned data, is:
互相关函数的长度为 2N-1，因此对齐数据的中心值为：
np.argmax(signal.correlate(data0.y, target.y)= N - 1 (zero-based)
np.argmax(signal.correlate(data0.y, target.y)= N - 1（从零开始）
shift = ((N-1) - N) * dx= (-1) * dx, when we really want 0 * dx
shift = ((N-1) - N) * dx= (-1) * dx，当我们真的想要 0 * dx
