[ENH] Predictions: Show errors#6012
Conversation
b2695ef to
91a9cf6
Compare
|
I fully approve this functionality! It enables selecting the most underpredicted, overpredicted and best predicted instances and showing them in the scatter plot. I also like the many options for error computation, discretely placed at the top. |
91a9cf6 to
bf07d10
Compare
Codecov Report
@@ Coverage Diff @@
## master #6012 +/- ##
==========================================
+ Coverage 86.60% 86.61% +0.01%
==========================================
Files 315 315
Lines 67317 67479 +162
==========================================
+ Hits 58297 58447 +150
- Misses 9020 9032 +12 |
bf07d10 to
6994a85
Compare
6994a85 to
90db1b0
Compare
90db1b0 to
5dd7b1a
Compare
| elif self._probs[column].size: | ||
| return 1 - self._probs[column][row, int(actual)] | ||
| else: | ||
| diff = self._values[column // 2][row] - actual |
There was a problem hiding this comment.
column was already divided by 2 above (line 1271)!
I must say, there is quite a lot of %2==1 and //2 in this file... :)
There was a problem hiding this comment.
column was already divided by 2 above (line 1271)!
It is disturbing that (seemingly thorough) tests didn't catch this because they used only a single predictor... :(
I fixed this and modified the tests to use two predictors.
I must say, there is quite a lot of %2==1 and //2 in this file... :)
Apparently (at least) one too many. :)
5dd7b1a to
def1d61
Compare
Issue
Closes #5971.
Description of changes
Add columns with errors for each predictor.
Please comment on functionality and interface. Checking and tests will come afterwards.
Try it on housing and heart disease; errors on iris are too small.
Includes