Update forecasting tutorial using scan primitive

[fehiepsi]

Suggested by @neerajprad!

[neerajprad]

Thanks, this will be really useful as a reference example for everyone!

Just some minor nits:

, we have use condition handler here.

use --> used

The reason is we also want to use this model for forecasting. In forecasting, future values of y are non-observable, so obs=y[t] will become invalid.

"will become invalid" - would suggest making it more specific, i.e. does this raise an index out of bounds error?

Using condition, when the length of scan is larger than the length of the conditioned/observed site, non-observed values will be sampled from the distribution of that site.

unobserved values?

[neerajprad]

On another note, I think we need an additional indent after doctest... for the scan documentation. I'm not sure if the doctest is running, but it is showing up in the documentation.

[fehiepsi]

Thanks Neeraj, I'll address them soon.

[fehiepsi]

@neerajprad I just incorporate your suggestions.