I have many sets of univariate sample data as the following: 
$$
X = \{x_i, i=1,2,...N\} 
$$
and I'm using Parzen-window density estimation with Gaussian kernel. 
$$
\frac{1}{N}\sum_i^N \frac{1}{\sqrt{2\pi h^2}}exp(-\frac{(x-x_i)^2}{2h^2})
$$
But I'm not sure how to choose the window size h.

In https://www.cs.utah.edu/~suyash/Dissertation_html/node11.html, it seems that the sample standard deviation was used as h. Namely, `h = std(x)`.
Is this a right way to choose parzen window size?

If it is not a proper way, then is there any other data-driven way to get the h?