Differences I've noticed between Anki and Pleco SM2 profile

CalebLA

秀才
Hello,

I’ve been using the Pleco SM2 profile (Anki emulation profile) for over two weeks now, and have been comparing it to my Anki deck.

I noticed some differences and am not sure why the Pleco deck is behaving the way it is.

1) The Anki Settings I am using worked very well for me (I found them at Refold.la Basic Anki Setup). I decided to migrate these settings over to Pleco. Even though some settings didn’t share the same names, I assumed (and I may be wrong) that settings like “Interval Modifier” (Anki) and “Interval Scale” (Pleco) were the same. I have taken a screenshot of my settings so you can compare.

Interestingly, when I would start a test, the new cards in Anki would go from <10 min to 24 hr, then 3 days, then 8 days, etc; whereas the Pleco deck would go from <10 min straight to 2 days and 16 hrs (most values varied around this number), and then 4d (n) hrs, and so forth. It would only go to 24 hr if “Interval Scale” was set to 100%.

Seeing that the intervals didn’t match my Anki deck, I added “steps” of 24 hr, 3d, 8d to Pleco so it would be similar to Anki. This seems to have worked okay, although sometimes I see odd numbers like 3d 5hr (or similar). I am wondering why Pleco’s algorithm is not the same as Anki right off the bat (with the same setting values). FYI, I only use the “good” and “again” buttons.

2) In Anki, no matter when it is that you will see a card again (let’s say 27d), if you hit “again” it shows the card will be seen in <10 min. With Pleco, most “again” buttons show <60 sec (new cards) or <10min, but some show numbers greater than that. If the card shows >3d until next review, it is likely the again button will show a number greater or equal to 24hr, not <60 sec or <10min like it should. I adjusted the Lapse relearning steps, and the Lookshead values, but the problem remains.

3) Lastly, the card Lookahead hasn’t been working very well for me. Sometimes it simply removes the card for the day, even if “again” is pressed, or shows the same card immediately after “again” is pressed, and subsequently removes the card for the day. This defeats the purpose of trying to review that card over and over in the same session until I can hit “good” twice in a row, like I would always do in Anki.
If I don’t use lookahead, I might end the test, but some cards will not have gotten past the “learning” phase because I finished the test before the 10 minutes in which a learning card would appear again.
Furthermore, the next day, I will have 20+ cards sitting in the ‘learning’ stage, whereas with Anki you only have new cards and review cards - cards are added to the ‘learning’ pool only if you hit ‘again’.

Not sure if this is the default Pleco behavior, if these are bugs and other users are experiencing this, or if it's just me and I’m using the settings wrong.

Please let me know if you need any clarifications.

Thank you, and any help would be greatly appreciated!

Caleb

IMG_0034.PNGIMG_0035.PNG
 

mikelove

皇帝
Staff member
Thanks! This sounds like a couple of bugs, or just us misunderstanding something how their algorithm worked (I guess it proves that we didn't rip of any of Anki's code :) The intent was that the behavior would match Anki more-or-less exactly, since the main goal of this feature was to let people bring their Chinese decks from Anki into Pleco and continue studying them alongside anything new they added in Pleco.

To clarify on #1, were Pleco and Anki originally configured to the same steps? (both had 1m and then 10m)? So then you added steps to match the intervals you were seeing on the Anki side when Anki was out of 'learning' and engaged in regular algorithmic SRS?

Also, where are you getting these card interval numbers? Are they showing up in the top status fields, or are you getting them from Organize / Card Info, or from the next interval previews under the score buttons, or are you computing them from when the cards actually appear? Pleco - not totally unlike Anki - maintains an "algorithmic interval" but then it doesn't necessarily schedule cards based on it if they're in a phase where they're appearing in set steps instead of with SRS, so it's possible the issue in some cases might simply be that it's reporting the wrong type of interval. This issue #2 sounds very much like it might be that sort of problem, since it ought to be previewing the interval based on steps (in which case 'again' always = 10m).

With #3, there's a 'timeout' feature in some of our other test profiles that may be activating here by accident - I don't think we appreciated that lookahead served this function in Anki (we thought it was simply a way of dealing with imprecise intervals).
 

CalebLA

秀才
Hello Mike,

#1, Yes, Pleco and Anki were originally the same steps, and after that, I only changed the Pleco steps. Anki's steps are still at "1m 10m."

#2, I saw the card interval numbers on the buttons. However, I am pretty sure it accurately reports when the card will be seen again.
I took some screenshots and started to notice a pattern.
The again button seems to be reverting to the step before the current step for 'good'.
For example, on the screenshots, you can see a card shows "again“ at 24 hrs, while "good" is at 3d, and when "good" is at 8d, "again" shows 3d.
For new cards, it shows a "good" of 10 sec, and a "again" of 60 seconds, which makes sense following this pattern.
There are some strange outliers though, like the card that shows "3w2d" under "good", and 10 min under "again".

I just took a test, and I can describe in more detail what happened.

I started the test with 15 new cards, 17 in "learning" (which shouldn't be the case), and only 5 in "review".

I was tested on the new cards, which was fine. However, most of the 17 cards already in the 'learning' stage (not new cards added to the learning stage during the test) had "again" values of 24 hr. I didn't remember most of them and hit "again", and I realized they would enter into a vicious cycle and never get out of it until I hit the "good (3d)" button. But since those "again" values are set at 24 hrs, I won't see them again enough to be able to review them into my memory (like I would if the value was <10 min, and I could hit that over and over until I could hit "good" twice in a row).

Strangely, during the last 15 cards Pleco seemed to have a mind of its own.
I would press "again (<10 min)," and the card would be dropped from the number at the top. It kept going that way until the last card was reached, and then the test was over.
It makes sense that they were dropped from the pool because Pleco thought they would be seen in "10 minutes". Unfortunately, the test was ended by that time, and it did not give me any opportunity to learn those cards. FYI, the Lookahead value was set to the default 20 min.

So, just out of curiosity, twenty minutes later, I hit "start test" again. It gave me the "error, no cards to review" message.
I can almost guarantee all those 15 cards that were quickly discarded form my pool will be waiting for me tomorrow in the "learning" section.

Here are some screenshots for you.

Please let me know if you have any more questions!

Thank you,

Caleb

IMG_0043.PNGIMG_0037.PNGIMG_0038.PNGIMG_0041.PNGIMG_0036.PNG
 
Last edited:

mikelove

皇帝
Staff member
So I'm currently seeing a few problems on our end:

1) It seems like the interval previews reflect the random variation we add to algorithmic intervals (a variation that's also used in Anki). So when they seem a bit higher / lower than they should be, it's most likely on account of that; obviously the previews should ignore that random change as they do in Anki. (and indeed the random factor added to the preview might not even be the same one as is applied to the card on choosing that answer)

2) It seems like Anki's interval previews do not reflect the interval modifier, while ours do. We're consistent with them in that we don't actually *store* the modified interval on the card, but if you answer one of these cards, backtrack, then tap on the (i) button to bring up Card Info and go into the card's statistics, you should see a big disconnect between the "algorithmic interval" - the one actually stored on the card - which should be more-or-less consistent with what Anki would give you, and the "scheduled interval", which reflects that scaling factor.

The issues with lookahead and with 'again' intervals longer than 10m I'm still trying to reproduce - could you possibly email me (mikelove at pleco dot com) a backup of your flashcard database (Organize / Import/Export / Backup Database)?
 
Top