[BROWSER] Support new ordinal models, fix old issues & other minor stuff
What does the code in the MR do ?
BROWSER: Fix, feat & refact:
- Add support for ordinal model (expectation & normalized by max_level per feature by default)
- Fix behavior with univariate & linear* models
- Fix space-shift computation for logistic & logistic parallel models (--> the metric-related factor was not present in logistic parallel model, and was multiplied at the wrong place for logistic model)
- Clean-up code, update some js dependencies and drop outdated model examples (to avoid the need to maintain them in addition to all other model params example in the repo)
- Leaspy favicon
* at some point (~v1.0?), linear model also got v0 = exp(log_v0_estimated)
and so behavior was incorrect..
Where should the reviewer start ?
cf. updates in js files
How can the code be tested ?
Open the browser app, load your models (or the toy ones in tests/_data
, example/
or in leaspy/datasets/
for instance) & play with the app (individual parameters tweaking, personalization)
When is the MR due for? (review deadline)
1 week?
What issues are linked to the MR ?
After !78 (merged) browser app did not support ordinal models & during implementation I spotted some issues that I solved (cf. above)