orange3
orange3 copied to clipboard
PLS: Move from Orange-spectroscopy
Issue
Move PLS widget from Orange-spectroscopy.
Includes
- [X] Code changes
- [X] Tests
- [x] Documentation
Codecov Report
Merging #6734 (c21b95a) into master (88f1048) will increase coverage by
0.02%
. Report is 21 commits behind head on master. The diff coverage is96.19%
.
:exclamation: Current head c21b95a differs from pull request most recent head 0144062. Consider uploading reports for the commit 0144062 to get more accurate results
Additional details and impacted files
@@ Coverage Diff @@
## master #6734 +/- ##
==========================================
+ Coverage 88.14% 88.16% +0.02%
==========================================
Files 322 325 +3
Lines 70586 70773 +187
==========================================
+ Hits 62215 62398 +183
- Misses 8371 8375 +4
Thanks guys!!!
I particularly don't like how ellipsis is implemented because it is hardcoded into PLS and forces the first two components.
I saw that people wish for something similar also for PCA, and, also sometimes they are interested in other components, for example, 2nd vs 3rd.
If I understand the idea behind it, a simple multivariate (ok, 2-variate) distribution is fitted on the data set, and then the ellipse of a certain percentage can be drawn. The code by @VesnaT only needs two columns of data, and I doubt it requires any assumptions only covered by PLS. So it should be general.
I see two ways of implementing it:
- A separate widget that allows users to select two features and then adds the column.
- Add these (general) ellipses to the Scatter Plot.
I would go for the second option. I was also asked quite a few times why we don't have this. Also, these ellipses fit into scatter plot visualization at least as much as regression lines (and even better than those that do not treat variables as independent).
We discussed this and decided on adding the ellipse to the scatter plot widget.