Unable to remove old title training

problem
1

Hello Samuel,

I hope you have been well. Thank you for your continued development and support of my favorite RSS reader.

Sometime after your recent training interface overhaul, I began experiencing a problem removing old title training from feeds. If I add new title training to a feed, I can remove it without a problem. However, old training for titles seemingly cannot be removed. This problem does not seem to affect other training, like author or categories/tags.

This problem is occurring in the web interface. I am using Windows 11. I tried this on both Chrome and Pale Moon browsers. The results are the same. My account is at the Premium level.

Steps to reproduce:

1. Pick any feed

2. Right click → Train & filter

3. Find a title phrase that was trained prior to earlier this year

4. Click the thumbs up/down button to untrain the title, which turns gray

5. Click the “Save 1 classifier” button, which automatically closes the window

6. Right click the same feed → Train & filter

7. The title phrase which was just untrained is still present

One oddity is that if you change a thumbs up to down or vice versa, the change stays. It seems that only trying to remove the training is currently not functioning.

Side note: this problem also seems to affect previously removed feeds as well. I would like to go into the “Manage Training” interface to clear out feeds I am no longer subscribed to, but am not able to do so.

Please let me know if I can provide any additional information. Thank you!

pk

2

Thanks for the detailed report, PK1. I was able to reproduce this.

The issue is that older title classifiers in the database were created before some newer fields (scope, is_regex) were added to the classifier model. When you try to remove training, the lookup query includes those newer fields, which don’t exist on the old documents, so the database can’t find them to delete them. Changing from like to dislike appeared to work because it created a new classifier with the updated fields, but the old one was actually left behind.

I’ve pushed a fix that adds a fallback lookup without the newer fields when the primary lookup fails, so old classifiers can be found and removed properly. This will be deployed shortly.

3

Samuel,

Thank you for the update! I’m glad you were able to reproduce the problem, and find the source of the issue!

pk

1 Like