[FIX] owlearnerwidget: Fix output initialization #1562
Conversation
Always output the initial learner, disregarding the `auto_apply`.
Assert that initialized widgets have a valid learner output
ales-erjavec
force-pushed
the
base-learner-auto-commit-fix
branch
from
87ce67e
to
5c6b60a
Compare
Current coverage is 88.66% (diff: 100%)@@ master #1562 diff @@
==========================================
Files 78 78
Lines 8099 8124 +25
Methods 0 0
Messages 0 0
Branches 0 0
==========================================
+ Hits 7197 7203 +6
- Misses 902 921 +19
Partials 0 0
|
| @@ -123,7 +123,7 @@ def __init__(self): | |||
| self.preprocessors = None | |||
| self.outdated_settings = False | |||
| self.setup_layout() | |||
| QTimer.singleShot(0, self.apply) | |||
| QTimer.singleShot(0, self.update_learner) | |||
There was a problem hiding this comment.
Wouldn't it be safer to call self.unconditional_apply here? It doesn't change anything for the base class since update_model doesn't do anything without the data, but derived classes can add some code to the apply method or change update_model to do something without data.
There was a problem hiding this comment.
I did try that. At the least that breaks 'Linear Regression' widget because it overrides the add_bottom_buttons method and does not 'define' unconditional_apply so it fails to even initialize.
There was a problem hiding this comment.
I see. Still, I think it's too dangerous to skip the potentially overridden apply. It can lead to a situation in which the widget will, for instance, ignore some setting when force-applied.
Let's require all widgets derived from OWBaseLearner to have unconditional_apply. All that don't, need to be fixed. Linear regression, in particular, doesn't need to define commit at all (partially my fault :()
There was a problem hiding this comment.
@ales-erjavec, @astaric, please see #1601, which should also fix this problem. Two other widgets in education remove buttons, while (AFAIK) no widgets in core, bio, datafusion, network, prototypes or text extend BaseLearner.
What about replacing the above line with
QTimer.singleShot(0, getattr(self, "unconditional_apply", self.apply))?
If the widget does not have unconditional_apply it has either left apply as it is, hence calling it has the same effect as unconditional_apply would, or it has overridden apply, so we must call it anyway. If the widget creates its own Apply button with a callback with a different name, then it will have to reimplement the output initialization, too.
There was a problem hiding this comment.
'this problem' = the problem with linear regression, not the problem fixed by this PR here.
Always output the initial learner, disregarding the
auto_apply.