Reset parameters when dimension is changed

[lrennels]

This PR handles #898

[codecov]

Codecov Report

Merging #905 (095eea7) into master (aac4868) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #905      +/-   ##
==========================================
- Coverage   84.56%   84.56%   -0.01%     
==========================================
  Files          40       40              
  Lines        3882     3887       +5     
==========================================
+ Hits         3283     3287       +4     
- Misses        599      600       +1     

Flag	Coverage Δ
unittests	84.56% <100.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/core/dimensions.jl	89.41% <100.00%> (+0.52%)	⬆️
src/core/defcomp.jl	79.03% <0.00%> (-0.43%)	⬇️

---

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 aac4868...095eea7. Read the comment docs.

[lrennels]

A question here is what is the desired behavior?

When we reset :time, our solution is to map the original set parameter values to the array of new ones, and pad the rest with missing. The model is set to dirty, and upon rerun will pick up this new parameter Array. The model instance will of course hold old values until rerun. Ie.

m = Model()
set_dimension(m, :time, 2000:2005)
update_param(m, : mycomp =, :mypar, 1:6) #mypar has dimension :time
run(m)
m[:mycomp, :mypar]

# prints [1,2,3,4,5,6]

set_dimension!(m, :time, 1998:2005)
run(m)
m[:mycomp, :mypar]

# prints [missing, missing, 1,2,3,4,5,6]

Now taking a case like redefining the country dimension from ["USA", "MDV", "ALB"] to ["USA", "MDV"]. We have no promise that the new set of strings will be a subset of the old, so we cannot as easily use matching to clean up the existing parameters. One could say set_dimension!(:country, ["X", "Y", "Z"]) ... length hasn't changed but labels have.

Some ideas for if you change a (nontime) dimension are:

1. parameters with that dimension are completely reset to nothing parameters and the user must update them again
2. if the size of the dimension has changed, the array parameters with that dimension are reset to all missing
3. don't automatically resize anything and assume the user will call update_param! ... but if they don't then the problem above will persist ... if they don't call update_param! which is pretty weird in terms of sizing etc.
4. warnings of some sort

In this PR I implement (2), which is pretty consistent with how we treat time, but if we remove the _resize_parameters! part it will implement (3) simply by getting rid of the copyto! ... side note is this bad for speed? We use copyto! in updating the model instance and it seems fine to leave it there, where we are assuming parameter sizes don't change since dimensions don't change, thus this wouldn't impact Monte Carlo performance.

[lrennels]

Will remove the warning, useful for testing.

[lrennels]

Update: we want this option: parameters with that dimension are completely reset to nothing parameters and the user must update them again