tidy.mvgam() WIP
#100
tidy.mvgam() WIP
#100
Conversation
There was a problem hiding this comment.
Thanks very much for this initial effort @swpease. I'm hugely appreciative, and overall I think this covers most of what users would like
|
Alright, I think it's mostly there now. I still want to split out the different covariance matrixes for heirarchical correlation cases, but think it warrants having a new The best tests I could think of were just some snapshot tests. I opted for this unappealing JSON output because the regular output (a print) was giving technically-not-false errors due to varying output truncations. I can't say I've ever had seamless experiences with the testthat snapshotting capabilities. |
`tidy.mvgam()` snapshot value tests The snapshots (note: not snapshot values) record what's printed, and the `check()` call led to a different truncation compared to `test_active_file()`, so I'll just use `expect_snapshot_value()` instead, even though it's harder to read. `tidy.mvgam()` snapshots tests covers the majority of main use cases, but is missing a random effects test and a heirarchical correlation test. `tidy.mvgam()` docs; term name Wrote documentation for method. Also settled on names for the "term" column contents. `tidy.mvgam()` trend formula w/o trend model These models have error terms (sigmas) which were not being included. `tidy.mvgam()` random effect specific betas Decided it would be better to include these than not; can be easily filtered out if undesired.
- added a seed to all mvgam_example*'s - added in a new example, mvgam_example6, for an AR with heirarchical correlated residuals model - added a snapshot for tidy.mvgam for this example - Implemented `tidy.mvgam()` Sigma splitting
|
I implemented the Sigma splitting and added in a new "mvgam_example" (#6, for heirarchical correlated residuals models). I filtered the Sigmas by non-zero variance, which I believe does the job -- I had initially been going down the indexing tetris route. |
draft method
Some comments on the current status:
gr, the Sigma matrix should be split out by group, but haven't implemented that yet.mvgam_example[n]s cover a good portion of possible outputs, but this method is supposed to handle every type of model. However, it seems excessive and minimally useful to cover that whole range with additionalmvgam_examples.Sorry this took to long -- I wanted to go over all the model types, and I've been managing eye strain so I can't just stare at text on my monitor for hours on end.