Combined g3_param_table / g3l_random_dnorm()

[lentinj]

We have random deviates with g3_param_table() & g3l_random_dnorm(), need to expand this for a projection.

• Create a g3s_modeltime array of values
• Populate initial selection with g3_param_table
• Populate gaps with rnorm(), based on g3l_random_dnorm() settings
• Use that instead of g3_param_table()

See:

• https://github.com/vbartolino/g3_GoB/blob/4e38a56e360c91ae05cbc2c32ec86d1d94b23b55/vendace_g3/gadget3/09-SRR-RND/setup_model.R#L151-L155
• https://github.com/vbartolino/g3_GoB/blob/4e38a56e360c91ae05cbc2c32ec86d1d94b23b55/vendace_g3/gadget3/09-SRR-RND/setup-random.R#L5-L14

[lentinj]

@bthe / @vbartolino for reporting of values, would it be better to have a combined array of random deviates & projected deviates, or just projected values and rely on parameter table for former? I'm assuming a combined array is more useful, but would be good to know if you think otherwise.

[bthe]

Would this mean that the values for the random effects would not appear in the parameter table?

[vbartolino]

does one option exclude the other? if possible I think it would be useful to have both and in any case intuitively a user would expect to find the parameters of the random effect in the parameter table

[lentinj]

Yes, the random effects will be in the parameter table regardless, having them in the report array won't affect that.

This is mostly a question of how to structure the array internally. If we want a report to include the random effects before projection, then I'll arrange to code so that happens by default.

[bthe]

OK, then I would think that most logical thing would be, as you suggest, to have both the estimated random effect and projected values.