Fitting stan model to 1 wave

[joshuategegne1]

Hi Shozen,

We have tried to use the run-stan-single.R script to fit the model in hsgp-eq-cd.stan to 1 wave of the covimod data, and have had a few issues (mainly with some of the helper functions) that may be of your interest.

On line 71, there is an error when applying add_contact_vector, this is because the third argument specifies if you're working with 1 wave, so single must be set to TRUE. So dt.offsets is not necessary

On line 77, when running add_row_major_idx, I found that dt.offsets didn't have all the variables required but dt.cnt did, so we changed the 2nd argument. I also noticed that the u column (index for waves and repetition) was missing but I can see you have made a note of this in stan-utility.R.

On line 80, when using add_part_offsets the second argument is contacts, so rather than being set to dt.offsets, this should be set to dt.cnt

On line 95, when applying add_hsgp_parms, args$hsgp_m needs to be specified as the 4th argument, the second dimension of the HSGP basis approximation

After changing this, I was able to compile the model but then get an error when fitting the stan_data:
Error: Missing input data for the following data variables: N_MM, N_FF, N_MF, N_FM, ROW_MAJOR_IDX_MM, ROW_MAJOR_IDX_FF, ROW_MAJOR_IDX_MF, ROW_MAJOR_IDX_FM, log_N_M, log_N_F, log_S_M, log_S_F, log_P_M, log_P_F.

With these constants and offsets that are missing, you can ensure that they are added to stan_data by changing survey = "COVIMOD" to survey = "POLYMOD" in the helper functions from stan-utility.R. After making these changes, I got a new error that there was a dimension mismatch in log_N_M (it is meant to be length 85 but is actually smaller)

Thanks,
Joshua

[ssl219]

Dear Shozen,

I hope you are well.

Following Joshua’s issue, just two more things to add:

1. On line 71 of the run-stan-single.R script, for a single wave on the COVIMOD study, the dt.offsets argument was not needed as outlined in the definition of the add_contact_vector function in the file stan-utility.R :

instead of:
stan_data <- add_contact_vector(stan_data, dt.cnt, dt.offsets)

we used:
stan_data <- add_contact_vector(stan_data, dt.cnt, single=TRUE)

2. We have also tried fitting the hsgp-eq-lrd-a65 model and we cannot seem to find the required helpers.stan file in the Github repository? The helpers.stan file is mentioned in line 4 of the hsgp-eq-lrd-a65.stan file.

Thank you for your time and help,
Selena Linden