2.0.8 Beta 2 Bug Report / Feedback Thread

[anchan42]

anchan42 - hmm... you're still using the dictionary entry remapping feature, right? ("Force defns from dicts") I just double-checked the code for that and I don't see any obvious sources of problems - could you give me some specific examples of flashcards whose definitions don't update correctly when you backtrack to them, along with the names of the dictionaries they normally link to and the current list of dictionaries selected for remapping? (and whether or not "Use custom card defns" / "as a fallback" are checked?)

To be honest, the flash cards is a bit over engineered for me (in other word, I am too dumb to use it properly :roll: ) I obviously had ticked the "Force defns from dicts" option by mistake. "Use custom card defns" / "as a fallback" ware not checked and only TL dictionary was selected.

After I ticked the "Force defns from dicts" option off, the back button seems to work fine.

 

[mikelove]

mfcb / HW60 - reproduced and fixed the Unihan compounds issue, thanks!

HW60 - are you sure you were exporting those as XML and not text? It's actually normal that cards are repeated in text exports in 2.0 - duplicating them is the only way we can get cards that belong to multiple categories to show up in all of those categories.

If you were getting this behavior even exporting as XML, how exactly did you export those cards, using a "Batch" command or something else?

mfcb - full-text search still isn't supported in user dictionaries, and won't be until we drop support for Palm OS. (but should come pretty soon after that, though it'll be optional even then since it's extremely storage-space-intensive)

anchan42 - it's over-engineered for a lot of people, actually Anyway, reproduced the bug now, so thanks for mentioning it again!

 

[HW60]

HW60 - are you sure you were exporting those as XML and not text? It's actually normal that cards are repeated in text exports in 2.0 - duplicating them is the only way we can get cards that belong to multiple categories to show up in all of those categories.

If you were getting this behavior even exporting as XML, how exactly did you export those cards, using a "Batch" command or something else?

I took Flashcards / Export with cards in categories, only 1 category (with the 6 model flashcards), File Format XML, Include categories, statistics and card definitions. I think I know what happened: As I just wanted to test the Card Filter "Not correct in row", I created 6 cards using the "+" button in main screen. Usually I convert all cards into custom cards and delete the Trad. information, but not for this single category. Therefore the exported cards also have the Trad. information in a separate record / line in Excel.
I will not look at XML files again as soon as Beta 3 with the card history is available.
This version of Pleco is actually a great help for my chinese studies! Do you have a rough estimate as to when the character separation / radical separation I saw in the IPhone video will be implemented in Windows mobile - this is probably the last real improvement I would like to have?

 

[daniel123]

When using flashcard Repetition-spaced with scroing type "Automatic" instead of "Old Automatic" I wonder if there is any equivalent to old parameter: "Difficulty scaling: All Decreases"? I noticed that all wrong cards now will be set to score 100. I prefer the old behaviour where a card with high score not immediately decreases as fast.

Actually I do not understand the new Automatic. Maybe it could be done by Tweak Parameters.

 

[Alexis]

Alexis - that is intentional, yes; even though those cards were already reviewed, they're scheduled to come up sooner than any / most other cards, which is why they're coming up right away in the new session. If you'd like to skip them, though, just add a Card Filter "Time filter" for "last reviewed" "not within" "1" "days" (or "calendar days").

Thanks Mike, that did exactly what I wanted.

 

[daniel123]

"Yes" actually is supposed to review cards that aren't due yet and extend their due dates; No stops reviewing, and Cancel reviews them without extending their due dates. Are you seeing different behavior on your system? The scores might not change even with "Yes" if you've enabled "only change once per day" in Advanced Settings / Scoring, but the "last reviewed" date should be updated so that the interval until the next review will still be extended.

Indeed as far as I understand it I see differrent behavior. When saying "Yes" even cards that have been last reviewed in 2009 do not change their score to a higher value if answering correct. But if I anwser wrong the score will decrease to 100. And the review date changes in both cases.

But I thought "extending due dates" (in this case saying "yes") would mean changing score (positvely) whereas "not extending" (in this case "cancel" what I cannot do because of my German version of WM) would mean not extending score because due date is related to score. Maybe I misunderstand it.

 

[mfcb]

in my case (repetition spaced, manual scoring) everything works as expected. if i answer "yes", all cards answered correct increase their score by 15%, cards answered incorrect decrease score by 25%... that the review date is updated i just trust mike to be true, hehe... never tried the "cancel" option (i updated my phone with an english rom), i'm pretty sure, even i have that button, i will never press it...

 

[daniel123]

You are right. Thanx that brings us closer to the problem. Manual work correctly with increasing score in this case. Old Automatic the same. But Automatic that I usually use does no increase the score.

 

[mikelove]

HW60 - ah, that makes sense; those actually should have an extra line, just not an entire extra XML record with its own dates / scores / etc.

daniel123 - yes, there is an equivalent; in "Tweak Settings" set "Scale score dec" to the percent you want the score to be scaled to for an answer of 1/2/3. The default "0" just drops the card back to the "minimum score," but if you set the value for a particular answer score to, say, 20%, the score will be set to 20% of its previous value.

The due date is actually a combination of the last date reviewed and the score; the score determines the interval that needs to pass since the last date reviewed before the card will be reviewed again. So you extend the next review date just by changing the last review date. The "Cancel" mode doesn't even change that - it keeps both that date and the score unchanged - while the "Yes" mode updates the last reviewed date but only changes the score if it's supposed to anyway.

So with the new "if rep-spaced is early" option set to its default "don't change," that means the score won't update at all if a card is reviewed early, even in that Yes mode. "scale expontentially" would be a good choice there if you still want the scores to count for something but not for much; it squares the portion of the actual interval that passed, then multiplies the score change it would otherwise do by that. So if you answer a card correctly and only 1/2 of the target interval has passed since the last review, the score will increase by (1/2 x 1/2) = 1/4 of the amount it would otherwise.

It's looking increasingly likely that we'll drop support for "frequency-adjusted" study altogether in a future release (maybe 2.1), since that will allow us to abolish this confusing Score business and just describe everything in terms of nice simple review dates / review intervals.

 

[sfrrr]

Mike Love said:
It's looking increasingly likely that we'll drop support for "frequency-adjusted" study altogether in a future release (maybe 2.1), since that will allow us to abolish this confusing Score business and just describe everything in terms of nice simple review dates / review intervals.

Oh, be still my heart. I simply can't figure out the niceties of the current system.

 

[mikelove]

Nor can a lot of people - I kind of wish we hadn't even promised flashcards on iPhone, actually; that way, we could wait to release them until later in the year when we've done some redesigning, and start off with a cleaned-up system on iPhone from the get-go. As-is, we're racing to get our current system working on iPhone only to have to junk / replace a significant portion of it soon after.

 

[daniel123]

I like the new scoring feature very much. Now I have the possibility to change everything as I want. Everything feels positive.

In my opinion no need to worry about that it is too complicated. You can give a few examples in the documentation in which case to change which value.

The people who are new to this flashcard system do not need to change anything and can use the default values.

The longer somebody studies flashcards the deeper will the understanding of the parameters.

I think at the beginning it looks more difficult than it really is.

I like to have a score to get an idea how good I know any card. IMHO this has much more power and comparable meaning than instead thinking in repetition dates as measure value.

Additionally this new algorithm is based on all of your experience and the wishes of forum users. Now it gives everybody great possibilities to custom it as he/she likes. A new calculation system would mean for you and us a lot of time to reach this actual level again. AFAIK there are things that are more important.

That's why I hope you keep this caclulation system for a while.

I think Pleco's actual flashcard system works great now. I use frequency-adjust to study small lists with new vocabulary and repitition-spaced to handle the big list of all existing cards.

Because flashcards are working so good now, I have the feeling that the only thing that is now missing are improved sound capabilities (for example pronouncing of words that have 儿 at their end correctly as one syllable instead of two, not stopping after 8 syllables, smaller breaks between syllable pronunciation or recognition of words like inside the reader, etc.)

Anyway, looking at this release I once again have to say: great job Mike!

 

[mikelove]

Thanks. I don't think we'd be changing the underlying algorithms, we'd just be using more logical labels than "score" - if we talk about repetition intervals it'll be much easier for people to understand, a lot of people don't even know that Pleco's flashcard system supports SRS as things stand now. We might keep the option to still display "scores" though.

 

[marsch]

Thanks for the option to tweak parameters! Very useful. However, it doesn't quite run as I expect...
I only use Correct/Wrong, so I'm only interested in the 2/6 params. My settings:
Difficulty change: default; initial score: 6/C: 200, scale score inc %: 90,100,100, dec% 25,50,75, min/max score: 100, 73000, diff min/max: 50,200, divisor: 50
This runs as expected for score increases, but for decreases it doesn't seem to consider difficulties, but just divides the score by 2.
Then, if I change scale score inc% to 90,100,200, diff divisor to 100 - which i thought would yield the same result for correct answers - i get results that I don't understand,
e.g. score 1797, diff 112, correct answer. expected new score: 4025, real score: 2227, difficulty unchanged
with a wrong answer, the score again is just divided by 2 to 898, difficulty set to 110 (??? this seems to happen no matter what the previous score was).
So I guess I'll have to work with "Old automatic" (which works great) until I understand "tweaking" or until the next bugfix?

 

[mikelove]

Those inc / dec %s are score-dependent too - changing "score inc %" to 90 / 100 / 200 doubled the increase for correct answers. In your particular case, the smaller score increase might have happened because you were reviewing the card before it was actually due - are you using repetition-spaced mode? I'm not sure why the difficulty would have only dropped to 110 from 112, though.

Decreases are no longer scaled by difficulty with the new algorithm; we think the difficulty is already being factored in more than enough without it, since it means a more rapid increase in the card's score once you start answering it correctly again. Answering a card incorrectly that you've supposedly learned very well is a big deal, which is why by default the score drops all the way back to 100; if you've gotten that card incorrect then we really want to show it to you again right away to make sure that you learn it correctly again. With the new "Undo" function it's easy to go back and undo an incorrect answer if you marked a card incorrect by mistake, so now that we have that we think it's OK to be a little more aggressive in penalizing you for incorrect answers.

It wouldn't be difficult to bring back difficulty-scaling for score decreases, though, since we already have the relevant code working in Old Automated - just an extra checkbox on the tweak settings screen. Does anyone else want to chime in in support of that?

 

[marsch]

Yes, I was aware that doubling score inc% would double the increase for correct answer, but I thought by also doubling the difficulty divisor the result should still be the same??
I am using repetition-spaced, but those cards actually were due already. Hmm, still not sure what I did wrong...
Don't mind too much about the difficulty-scaling for decreases, although I'd prefer if "tweaking" was really "tweaking" the old algorithm, not something separate, meaning I could achieve the old behavior by choosing the right params. Maybe even an option to fill the tweaking dialog with the defaults from the "old algorithm"'s aggressiveness levels?

 

[mikelove]

Oh, now I see the problem - didn't notice your comment about the difficulty divisor being increased to 100. It's a divisor, so increasing it -> effective difficulty is lower. The difficulty can't be lower than 1.1 x the divisor (that's a hard limit we set to make sure cards' scores still increase at least a little bit), which is why your difficulty wouldn't go below 110 after a wrong answer. But the score was actually calculated perfectly: 112 / 100 -> a 12% score increase, double that since you're scaling increases by 200% -> a 24% increase, add 24% to 1797 and you get 2228.

 

[marsch]

I get it then. Thanks
Should've noticed that it's the score *increase* that gets scaled. I assumed it was the score itself...
Alrite, will continue tweaking then.

 

[HW60]

Full text search does not work with user dicts.

 

[marsch]

Re new syllable parser. I tried to look up "shuangfang" 双方 today (by pinyin). As soon as I type the "f" it goes back to 书. Old parsing doesn't work either.