[BUG]: Parameter matrix generated by stratifying certain model transitions are wrong

[liunelson]

We have this SEIRDH model where the infection-exposure process is S -> E controlled by I

\frac{d S}{d t} = - \beta S I
\frac{d E}{d t} = \beta S I - r_{EI} E
\frac{d I}{d t} = r_{EI} E - r_{IR} p_{IR} I - r_{IH} p_{IH} I
\frac{d H}{d t} = r_{IH} p_{IH} I - r_{HR} p_{HR} H - r_{HD} p_{HD} H
\frac{d R}{d t} = r_{IR} p_{IR} I + r_{HR} p_{HR} H
\frac{d D}{d t} = r_{HD} p_{HD} H

SEIRHD_base.json

When this model is stratified, we get this strange 2 x 4 transition matrix between S and E, wherein the S_u, S_v states appear on both the rows and columns. This doesn't sense.

• The transitions t0_v_u and t0_u_v are misplaced since they do not have S_u, S_v as output states
• If the rows are input states and the columns are output states, then t0_u_u and t0_u_v should be grouped; similarly for t0_v_v, t0_v_u (second index is for vax status of the I)
• This is suboptimal since it'd have two entries per matrix element
• Ideally, the column would use "hierarchical indexing" as in Pandas (https://pandas.pydata.org/docs/user_guide/advanced.html) where the first level is the output state combinations and the second level is the controller state combinations

This bug changes when I toggle off the "Create new transitions between strata":

• the transition matrix becomes 2 x 2 and the t0_u_u, t0_v_v transitions are no longer shown 👎

[liunelson]

This is probably a bug that @mwdchang is most familiar?

[shawnyama]

After digging around I found that there are two templates of the same name generated by MIRA. These are the templates in the mmt:
miratemplates.json

Here are the ones that are the same in the above file

Looking at the expanded view in the model diagram also shows that those connections do exist. The first pair is an S-> E transition which isn't expected. The second pair is S -> S which is expected.

[shawnyama]

Model from equations: sierdh.json
Stratified model: sierdh_strat.json

[mwdchang]

The duplicated template names aren't handled in our collapseTemplates function, but more than that the resulting AMR has more than one expressions point at a singular target. I filed a ticket with MIRA:

gyorilab/mira#348

[mwdchang]

This seemed okay after the MIRA fix.