search

thiswillbeyourgithub / AnnA_Anki_neuronal_Appendix

Using machine learning on your anki collection to enhance the scheduling via semantic clustering and semantic similarity
GNU General Public License v3.0
62 stars 1 forks source link

Move the tag ignoring settings to a user-customizable place? #9

Closed aleksejrs closed 2 years ago

aleksejrs commented 2 years ago

Anki adds "leech" to difficult cards. MorphMan adds lots of tags like "mm_comprehension" or "mm_fresh". Tags like "relation-blabla": "relation" needs to be ignored. I have some custom tags like "important", "urgent", "morphman_important" or "morphman_urgent".

thiswillbeyourgithub commented 2 years ago

Good suggestion, added in 2e6892a89adbda620ebe4dee461385c6c6ddb20a

I don't have the time to test it so I recklessly pushed it. I would greatly appreciate it if you could reopen this issue if you find an issue with the commit :)

aleksejrs commented 2 years ago

I guess it works, thanks. Prefixes would also be useful though, and the word "relation" only appears after splitting.

thiswillbeyourgithub commented 2 years ago

What do you mean ?

aleksejrs commented 2 years ago

Prefixes: tag.startswith(("mm_", "morphman-")) relations: Bury related notes https://github.com/Arthur-Milchior/anki-relation

aleksejrs commented 2 years ago

So cards sharing the same string after "relation-" are very related (like siblings, but not in a way Anki would know), but sharing just the word "relation" is nothing but "I have that kind of siblings".

thiswillbeyourgithub commented 2 years ago

I don't think handling "relations-XXX" is a good idea.

The whole purpose of AnnA is to find which cards are related automatically. If you use AnnA with low enough settings (say 50% of daily reviews(, and 80% of daily learnings if over multiple days)) it should actually never be a problem.

If two cards are very related but share absolutely no words, phrasing or tags in common then I think you're deck is not very well organized.

Regarding prefixes: I don't understand what you mean

aleksejrs commented 2 years ago

If you use AnnA with low enough settings (say 50% of daily reviews(, and 80% of daily learnings if over multiple days)) it should actually never be a problem.

I don't have a significant backlog now, so I don't really understand.

If two cards are very related but share absolutely no words, phrasing or tags in common then I think you're deck is not very well organized.

Conclusion two paragraphs below: okay, that doesn't seem significant.

The relations appeared before AnnA, and are used for new cards, like PushParadox does. It's not for fast learning, but it reduces the chances of the AnnA-treated problem appearing in the days between reviews of young cards. Another add-on like that is Bury Cousins (related cards that aren't siblings)

The potential problem is that — if AnnA's effect can be positive, can it also be negative (e.g., if I reviewed many important cards on one day, it might try hiding important cards from me the next day, just because they all have "important" and "frequency" tags)? — the meta tags may make a significant part of those words, for example in Image Occlusion or image-name cards. An image-name card is often just an image (1), two words (3), a year (4), a tag (6), and a relation tag (8) and the word "relation" is 1/8 of all words. An image-age card may be missing a name, but then "relation" can actually help.

Regarding prefixes: I don't understand what you mean

Okay, I don't really care if you implement that. I have 9 meta tags starting with "mm_", 3 tags starting with "morphman-", and many "todo-" tags. I can, of course, add the two former groups manually without a problem, and the third one is pretty much obsolete and is really supposed to be temporary or for suspended cards.

thiswillbeyourgithub commented 2 years ago

The potential problem is that — if AnnA's effect can be positive, can it also be negative (e.g., if I reviewed many important cards on one day, it might try hiding important cards from me the next day, just because they all have "important" and "frequency" tags)? — the meta tags may make a significant part of those words, for example in Image Occlusion or image-name cards. An image-name card is often just an image (1), two words (3), a year (4), a tag (6), and a relation tag (8) and the word "relation" is 1/8 of all words. An image-age card may be missing a name, but then "relation" can actually help.

It's actually not true. Tags are written once whereas fields of your cards are copied as many times as you want.

Hence, if you feel like your image occlusion cards don't have that many words, you can just increase the relative importance of those fields anyway.