Closed janosg closed 2 years ago
:exclamation: No coverage uploaded for pull request base (
main@9e7c32a
). Click here to learn what that means. The diff coverage isn/a
.
@@ Coverage Diff @@
## main #66 +/- ##
=======================================
Coverage ? 93.74%
=======================================
Files ? 34
Lines ? 2510
Branches ? 0
=======================================
Hits ? 2353
Misses ? 157
Partials ? 0
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update 9e7c32a...2e66846. Read the comment docs.
Current situation
There are only latent factors. Observed controls can only enter the measurement equations.
Desired situation
Observed variables can enter the transition equation.
Why not just fix parameters
There are a few important conceptual differences between observed factors and "fake latent factors" that have just one measurement with fixed loading and intercept in each period.
User interface
The model dictionary now can contain an entry called "observed_factors" which is simply a list with variable names in the empirical dataset. Those variables cannot contain missings. The set of observed factors is assumed to be same in all periods. If that is not wanted, coefficients can be fixed to zero.
The observed factors will enter the right hand side of transition equations in exactly the same way as latent factors. However, they do not have their own transition equations.
To-Do
model["labels"]
needs to contain "latent_factors", "observed_factors" and "all_factors" instead of just factorsmodel["labels"]["factors"]
now needs to use the correct one of the three factor entriesparse_params
,params_index
andconstraints
process_data
visualize_transition_equations
simulate_dataset