In the Inference of a SDE section og the Turing tutorials the noise added to the states are described in the noise model with the \phi parameters. Is \sigma supposed to capture observation noise? What is its true value?
odedata = Array(sol1) + 0.8 * randn(size(Array(sol1)))
0.8. That one needs a lot longer of a chain to converge though.
Thank you so much for clarifying! How does one get to 0.8 though?
I set that when generating the data. In reality, it’s measurement noise.
Ok thank you!
I am a little bit confused since in the mentioned SDE problem
data is used for inference and not
model = fitlv(data, prob_sde). So, in my understanding the data does not contain measurement noise.
I am asking, because I am trying to model a process with diagonal noise as SDE and in addition with measurement noise.
Yeah someone pointed out this typo yesterday, and I have a PR open for updating it. It should be
odedata is generated with measurement noise only, i.e. no noise model (no SDE) is used?!
Is this intentional or am I missing something?
I mean, you can do it with process noise too. What really matters is that it’s one trajectory of data.
Sure, but the expected solution of the sampling is then not close to
p = [1.5,1.0,3.0,1.0,0.1,0.1] but 0, or? This might be misleading in this set-up.
Yeah which is why using one output from the deterministic model (or mean of SDE outputs) is better for testing