Confusion about Confusion Matrix (pun intended)

wypee

Guys, please help me make sense of this performance vector. I know how to interpret the Accuracy tab:



But when I click on the precision tab, I get this:



Why the pred0,true0 cell count has gone from 1846 to 437? Why is the precision tab precision 19.14% not matching with any of the class precision on the accuracy tab (80.86% and 77.39%)?

wypee

Not sure why the images are coming as broken. Here are the links:

https://www.dropbox.com/s/0yjm2pdtw2a4zpm/1.png

https://www.dropbox.com/s/0btzchvmi1wi6iy/2.png

MariusHelf

This looks definitely like a bug. Can you please post a process that reproduces this behavior?

Best regards,
Marius

wypee

Thanks Marius. Here is the process:

https://www.dropbox.com/s/cf15krxbmllihlp/insult_process.rmp

This is for one of the kaggle competition to classify user comments as insulting/not-insulting. It uses the following sub-process:

https://www.dropbox.com/s/hzdj0ef7kczouol/prep_text_data.rmp

Regards,
Yogesh

MariusHelf

Thank you. If have created an internal bug report.

Marco_Boeck

Hi,

the issue has been fixed and will be included in the next release.

Regards,
Marco

DocMusher

Hi,
Perhaps a very stupid question, but the results presented in a confusion matrix are correct even with this bug (and can be used in scientific writings)?
Cheers
Sven