Closed CrtGrahonja closed 5 years ago
I assume the bulk of the problem will be solved via https://github.com/biolab/orange3/pull/3925. Discrete variables from Estimate Discrete and from Training are not the same. Before, one had to use Variable.make to make them the same. Will be fixed in the next release.
Also, remove the target variable from the Estimate Discrete. You don't need it. On this topics, there should be no crash when passing an empty target variable. Continued in #4060.
Describe the bug When trying to predict a class from text using Logistic regression (but any other regression as well) errors happen, as can be seen from the attached picture 'Errors':
This happens, when a Python Script is used to manipulate the input data. In the example a discretization of the target class has been done in the form of 2 classes: 1 (variable value 'je') when the original class variable was positive and 0 (variable value 'ni') when it's 0. As explained above, in the data to be estimated, the value was always '?'. As can be seen from the attached files and picture, the format of the variables is the same in both files.
This issue was not present when the Python script wasn't used (the branch "Estimate Continious").
To Reproduce Steps to reproduce the behavior:
Orange version: 3.22.0
Expected behavior Normal prediction executed like in the branch "Estimate Continious".
Screenshots
Operating system: Windows 7.
Additional context I attach the whole workflow, all three .TAB files used and the Python Script file. Prediction_TM_Errors.zip