gianlucadetommaso / stein-variational-samplers Goto Github PK

Never mind! The shape does not so depend on the observation, and both a=0.0 and 1.0 works similarly.
----EDIT-----
I want to reproduce your toy experiment for double banana in the paper in python. It seems like the shape of the posterior would heavily depend on your observation y. Could you share which value you take to get those figure? (In code it is just a random variable)

By the way, how is it possible to get a symmetric shape with $x_1 = 0$ in that example, if you take a = 1.0? Is it a typo that we instead should take a = 0.0?

I'm trying out the code in BOCPDviaSMC_Hawkes on a sensor dataset and am getting an error at this line rld = stats.rv_discrete( values = ( np.arange( min(t_+1, self.rmax) ), self.rlp ) ) in the class BOCPD. I printed out self.rlp :

Time: 6
[1.00000000e-02 1.81195171e-04 1.12306196e-02 1.08983524e-02
 1.72809904e-04 9.67517023e-01 0.00000000e+00]
Changepoint at time 6 for drastic decrease
Time: 7
[nan  0.  0.  0.  0.  0. nan  0.]

Any possible reason why I'd be getting nans (or any way to solve this issue)?

Thanks!

I believe that there are some files missing to run this. Specifically, I tried to run the double banana file (by first executing load_dir_double_banana.m and then running script_double_banana.m . It couldn't find the function SVN_H. I'm guessing that those files are in the folder "../samplers/stein", which isn't posted on the repository?