[FIX] Widget Logistic Regression: can handle unused values #2116
Conversation
| data = Table("iris") | ||
| data = data[:80, :] | ||
| self.send_signal("Data", data) | ||
| self.widget.apply_button.button.click() |
There was a problem hiding this comment.
This only checks that the widget does not crash. Could you also test that the output is sensible?
Can you try it with keeping the first and the third class, or the second and the third?
| @@ -92,10 +92,10 @@ def get_learner_parameters(self): | |||
| def create_coef_table(classifier): | |||
| i = classifier.intercept | |||
| c = classifier.coefficients | |||
| if len(classifier.domain.class_var.values) > 2: | |||
| values = classifier.domain.class_var.values | |||
| if c.shape[0] > 2: | |||
There was a problem hiding this comment.
Isn't this something different? Wasn't the previous code correct? Isn't classifier.coefficients equal to the number of attributes + 1? Or am I overlooking something?
There was a problem hiding this comment.
Previous code was probably not correct. c.shape[0] is not always equal to len(classifier.domain.class_var.values). That is in the case when we have a domain with more class values than contained in a data table. And of course always in the case when number of that values is 2.
|
Tests for this still do not pass. Have you looked into the issues caused by sgd reuse of the modified function? |
I can handle unused values and instead of writing "coef" or sth similar I write a second value name when only two values are considered. https://sentry.io/biolab/orange3/issues/222366525/ https://sentry.io/biolab/orange3/issues/208659372/
Test Logistic Regression: lint
Codecov Report
@@ Coverage Diff @@
## master #2116 +/- ##
==========================================
+ Coverage 71.04% 72.02% +0.97%
==========================================
Files 318 318
Lines 54563 54582 +19
==========================================
+ Hits 38767 39314 +547
+ Misses 15796 15268 -528Continue to review full report at Codecov.
|
Issue
I can handle unused values
and instead of writing "coef" or sth similar I write a second value name
when only two values are considered.
https://sentry.io/biolab/orange3/issues/222366525/
https://sentry.io/biolab/orange3/issues/208659372/
Description of changes
Includes