(updated) toy example where NUTS with dense mass matrix outperforms default NUTS

[Yuanjun Gao]

I just realized that I should use "thin=1" when using stan to check the autocorrelation. And I find that in my example, the dense matrix not only saves time but also get better sample result with higher quality(much smaller acf) The result is attached. Thank you for further comments.

Best,
Yuanjun