Linear Regression Widget (or more likely test learner widget) does not follow filtered attributes

[reportbug1]

Use case: Linear regression

• take a suitable dataset (more than 1 predictor variables, 1 response variable (to be predicted))
• do a regression with only 1 predictor variable
• do a regression with all the predicotor variables

in test learners they will show the same performance scores results (ie the results of the full regression, i suspect because of the data feed for test learners), so the performance scores of the regression with 1 only variable are false

as a check showing the regression statistics for each regression method does show the correct data

version 2.7.8

[ales-erjavec]

I suspect you are not using the widgets correctly (can you provide more details).

Providing input to a Learner widget (Linear Regression, Naive bayes, ...) only effects the Classifier/Predictor output not the Learner output. So if you are using Select Attributes -> Linear Regression -> Test Learners you should not expect that the selection in Select Attributes has any effect on the application of the learner in Test Learner widget. This might be confusing for the uninitiated, but this is by design. That is because the Learner widgets serve a double purpose:

• 1) Instantiating a Learner object with a set of parameters (resulting in the 'Learner' output)
• 2) Applying the learner (from 1) on the input dataset (resulting in the 'Classifier/Predictor' output)

[reportbug1]

but the learner object for regression also output the model so this model is ignored by "test learners"?

setup (works): data - XY data - lin regre - test learner (+ xy data as feed for test learner)

what does not (wrong output for the simple XY model): data ---------- lin regre - test learner -xy data-lin regre- (data as feed for test learner)