Open raindropsfromsky opened 4 years ago
🤔 I think I know where that is coming from. That's probably a mistake I made while I was debugging Qiqqa a while ago: I seem to recall hitting a serious performance drain with the "AI" in Qiqqa running on the lib under test and right now I'm not sure if I reverted the cutting back on that internal calc nightmare.
Anyway, needs investigation, as Qiqqa can do better in the tag suggestion department there. It's not brilliant, but auto-tags were a decent set for a machine to spit out, at least for my libs and (low) expectations then.
(There were odd bugs in the existing implementation though, already before I kicked it in the nadgers for performance reasons. I recall having had a look at how I could kill the duplicate "themes" discovered by Qiqqa under some circumstances, but at least I am sure I did not resolve that problem as I ran into some coding chunks which made this hard to fix. 😢
Needs follow up & analysis of the current state of affairs in v82pre9 -- I am assuming you're using the latest experimental Qiqqa release, right?
Yes, using the v82pre8
I ran the Expedition on a few files, and the result does not any sense at all.
For example, the 721-page Compendium file yields the following themes:
And the Supreme Court case file yields the following themes:
(I have shared both files.)
Finally, I ran expedition on the Qiqqa manual itself (just to see what happens). And this is the amazing result:![image](https://user-images.githubusercontent.com/9047168/77579709-c2605e00-6f00-11ea-82e1-f66b766642bb.png)
I see the following issues: