使用scipy.welch估算时间序列DF的功率谱密度

Question

我有一个带有datetime索引的pandas DF，其间距= 200ms，每个索引的对应值如图所示

print(filtered)

2016-07-14 16:31:19.000 -0.010054
2016-07-14 16:31:19.200 -0.011849
2016-07-14 16:31:19.400 -0.009564
2016-07-14 16:31:19.600 -0.001077

[20038 rows x 1 columns]

我想用scipy.welch函数计算功率谱密度。

f,pxx =welch(filtered.values.flatten(),5)        

但是当我运行这行代码时，功率密度数组pxx是nan

In [897]: pxx
Out[897]: 

array([ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan,

在时间序列数据帧上运行welch估计的正确方法是什么？在哪里可以找到导致welch函数输出nan的信息？

Answer 1

f,pxx =welch(filtered.values.flatten(),5)        

works fine on my side, make sure you have no missing values in your DF and your dtypes are correct (values are floats) first.

this should work

filtered = filtered.astype(float)
filtered = filtered.dropna()
f,pxx =welch(filtered.values.flatten(),5)