Getting started with spaced repetition

daal

探花
I would like to use spaced repetition to study my flashcards, but am not sure how to get started. I have added about 300 words so far altogether, sorted in various ways, and I would like to spend about 10-20 mins a day practicing them. Ideally, I would like to practice a set of about 30 or so words that would be repeated until I got them consistently right, and then as I learn these words, the set would be replenished with new words. I would also like to see a word again fairly soon after having gotten it wrong.

At the moment I am finding Pleco's options a bit overwhelming and have not been able to set up the flashcard system to my satisfaction. It seems that Pleco has a max # of cards setting, but that it does not appear to be available in the spaced repetition profile. This means that when I start a session, I have too many words and I don't get a chance to repeat the ones I got wrong. I could limit the words using categories, but then the words would not be replenished in a spaced repetition.

How should I set up my profile to get Pleco to show me a small pool of words fairly frequently, and to replenish that pool as the words get learned?
 

mikelove

皇帝
Staff member
1) Reset all of your card statistics, Organize Cards / Edit / Batch (towards the bottom).

2) In Card Selection, set "limit # of unlearned" to a fairly low limit.

3) In Commands, turn on "repeat incorrect" and configure it to your liking.

Pleco should now introduce new cards just a few at a time and only give you the rest of your words gradually as you master the ones you're already studying, and should review you on cards that you recently answered incorrectly later in the test.

We don't have a maximum # of cards setting in spaced repetition because it defeats the whole purpose of spaced repetition - if you're not reviewing cards at the proscribed intervals then you're liable to forget them. We are looking at adding an option for a set # of cards in a future release, but it would work by deliberately removing some cards from your review pool once you had too many so that SRS would be preserved.
 

daal

探花
Thanks for the prompt and helpful reply! I will get started with this setup and see how it goes.
 

daal

探花
This set up is going fairly well as far as maintaining a manageable number of cards goes. I did however find myself wanting to tweak the automatic scoring system, and am finding that a) it takes a lot of time to think about and b) doesn't respond to my tweaks in the way I had expected- in other words, I spent a lot of time thinking about how to tweak the scoring system, but it wasn't enough time to understand how to do it right.

There were two reasons that I wanted to tweak it. First, I had the feeling that I wanted to see the cards again sooner than the system was showing me, and second, the system didn't seem to be differentiating much between the three correct answer possibilities.

What I did was to lower the initial correct scores to 150, 250 and 400 for Qualities 4,5 and 6 respectively (btw, if they were referred to by the names that they use in the interface: "barely remembered," "remembered" and "remembered perfectly", it would reduce confusion), and I reduced the Correct scale score increase % to 50, 70 and 100 also for Qualities 4,5 and 6 respectively. I also raised the "card is learned if score >= " amount to 400.

I did this to one of the three scorefiles that I am currently using (one each for recognition, recall and writing), and the first thing I noticed was that the tweaks seem to be global. That is, when I went to tweak the other score files, I found that they too had already been changed to the values I set for the first. This was a bit surprising. The second thing I noticed was that the values I was seeing during the test didn't seem to jibe with the tweaks. The biggest surprise was what happened next. After doing the tests, the tweaks had mutated to new values. The first time I checked them, they had moved upwards a bit (except the card is learned value which presumably dropped, I didn't check), and the next time I checked they all had returned to their default values. o_O

This is probably a matter for customer support and not a public forum, but what I want to say is essentially that the system is too complicated for my liking. There are a lot of settings that affect when one sees the card next: If you answer right the easiness factor does something to the score change % and this gets multiplied by the current score, unless it was the first time you've seen a card in which case that number is overridden by the initial correct score. Or not. I probably don't understand it. In any case, I am stuck with these questions: which of these numbers should be tweaked, and by how much, and how to make the tweaks sticky? Although I do appreciate it being tweakable, I'd rather that it did not take more time to do so than studying Chinese. What I'd prefer to see as a user wanting to see the cards more frequently is a slider with the words "more frequent" and "less frequent."

By the way, don't offer to refund my money. Despite the settings overload, I am still finding it much more practical than Anki, and if I can't tweak it properly, then so be it. I do however wish to encourage the developer(s?) to keep people like me out from under the hood as much as possible.
 

mikelove

皇帝
Staff member
We're actually finalizing the new SRS configuration UI design right now - 'custom fields' ended up involving 80% of the work we needed to do for an SRS revamp anyway so we're getting a bunch of that stuff into our next update too (we think / hope, anyway). The new system has a sort of a 'progression' concept that should be a lot more intuitive than our current SRS scoring configuration (and a lot more powerful / flexible too, actually).

Re your current problem, though, 'scoring' is actually stored at a profile level, not a scorefile level. (the thinking being that people might like to use different scoring systems in different profiles) Is there any chance that that might also explain why the tweak values had changed? I can't think of any reason why the values would gradually creep back towards their defaults on their own, outside of some very bizarre / unlikely memory overrun or some such there isn't really any way that running a test could cause a change to those values.

If you simply want to adjust the gross intervals, though, the easiest way to do that would be to not touch Scoring at all and simply tweak the 'points per day' in Card Selection - that will quickly let you increase or decrease the spacing between reviews for all of your cards.
 

daal

探花
Tweaking the points per day does the trick. As with my initial question above, the answer is much simpler than the interface would have it appear. ("card selection >points per day" doesn't really indicate "reduce or increase spacing between reviews" to me...) I am quite looking forward to your new UI! As to the other problem, since the values have gone back to the default ones, and the reason for messing with them is gone, I'm not going to worry about it any more for now. Thanks for your help.
 

HW60

状元
We don't have a maximum # of cards setting in spaced repetition because it defeats the whole purpose of spaced repetition - if you're not reviewing cards at the proscribed intervals then you're liable to forget them. We are looking at adding an option for a set # of cards in a future release, but it would work by deliberately removing some cards from your review pool once you had too many so that SRS would be preserved.
I would very much like to have a fixed number of cards in SRS, and actually this has been discussed very often in this forum. There are many advantages
· you can choose the time you want to spend learning and repeating cards, e.g. "20-30 minutes a day" as @daal wrote
· you can skip a day repeating if you have no time
· you do not feel the increasing pressure of due cards that must be reviewed as otherwise you are lost because the amount of due cards steadily increases
· you can review a small number of cards on days where you would not start reviewing because of the many cards in "real SRS"

The system I want to propose is a spaced repetition system, but the user can choose the number of cards each session. The number of categories is changing automatically, as categories are "deliberately removed from the review pool", and some of the cards in some categories temporarily do not belong to the review pool. I made a small Excel sheet to show what I mean.
There are 30 cards in the pool, some old and overdue cards and most of them new cards, all in one category. The scoring system is very simple: score=100 when the answer was wrong, score: +100 when the answer was correct until a top score of 1000. There is a daily amount of 10 cards to be reviewed.

The card selection first takes cards from overdue cards, sorted by how overdue they are. If there are more overdue cards than cards to be reviewed, only the chosen maximum number of cards is selected. In the example there are 8 cards overdue which are all selected. Then there are 22 new cards, from which 2 cards are randomly selected. So on day 1 the desired number of 10 cards is reviewed, and the SRS pool has 10 cards. There are 5 wrong and five correct answers assumed.

On day 2 there are 5 more randomly chosen cards, together with the 5 cards with a wrong answer on day 1 they again make up for 10 cards. The SRS pool increases to 15 cards. Now on each day more cards from the category are randomly chosen and added to the SRS pool, if the number of due cards in the existing SRS pool is lower than 10 cards, the chosen maximum. On day 4 there are enough due cards for review, on day 5 there are 12 due cards, e.g. more due cards than to be reviewed. In that case 2 cards have to be eliminated from review - not from the SRS pool - and in the example the due cards with the highest score of 400 were selected (card number 24 and 30). These 2 cards are selected the next day because these are the only overdue cards then.
Day 8 is a free day without review. That leads to 15 due cards on day 9, from which 5 cards are eliminated, again by highest score. On day 12 there are only 9 cards in the category which need a review. To reach the desired number of 10 cards the next category has to be added to the potential SRS pool (the categories from which cards are selected for the SRS pool).

That's all. If there is no natural order of categories, the user has to be asked which category should come next, or he can define some order for his categories. Categories are not only added to the potential SRS pool, but can also be automatically removed. That is the case when the desired amount of cards to be reviewed is already reached without the category. Categories are added and removed last in - first out, the last added category would be the first category to be removed.
Some more settings have to be made, e.g. when should a card be automatically be withdrawn from the SRS pool? In the example 100 days overdue (card number 23) will do - this card does not belong to the SRS pool at the beginning (in that case the score of 100 of the card leads to an immediate review too).

Maybe the system looks complicated, and maybe it is complicated for the programmer. But it is very very easy for the user: just the desired number of cards and the desired category to start with - nothing more to think about.
There are more advantages than just the easy manageable workload.
· you can start again reviewing cards after a break of 10, 20, 100 or more days without reviewing which is almost impossible with the current system in Pleco.
· You can review a list of 1000 or 3000 cards with a constant daily effort - the actual limit new cards option does not keep your daily work constant

So you have a maximum # of cards setting within a spaced repetition system.
 

Attachments

  • Spaced_Repetition.pdf
    240.2 KB · Views: 701

mikelove

皇帝
Staff member
This is interesting, but my inclination would be to make removal a bit more obvious + flexible - provide a way to tell the system about your study priorities and then have the system respond by making it clear which priorities you're working towards now / which you're staying up on / which you're falling behind on / etc; if you only have time to study 100 cards today then you'll see a big angry red indicator that you're falling behind on priority X, and a few days or weeks after that you'll find that priority X has been disabled from active study entirely until you're ready to start up on it again.
 

HW60

状元
I wanted to propose a simple to use system. Many users would be happy to make their priorities clear by only choosing the maximum number of cards and the categories to select cards from. Of course I would appreciate statistics how many overdue cards I have in all of my flashcards and how that figure changes over time, but that could be provided for the actual SRS system already.

There are 2 different goals you can achieve with the proposed system:
- for users without overdue cards you can still adjust their daily workload and warn them with a colored monster how far they have fallen behind if at all
- for users who have fallen behind so much that they do not know how much and how to recover, you could give them a method to reduce the number of categories and the SRS pool and then gradually increase this pool again as described above.
Pleco could remove cards from the SRS pool which are overdue for more than e.g. 100 days and remove categories without cards in the SRS pool afterwards. Of course that should happen very transparently. You could ask the user before. But if that "big angry red indicator" would already exist in the actual system, more users than you might think about might find it in their deck.
 

mikelove

皇帝
Staff member
Simplicity is good, yes, my concern is simply that without obvious groupings users won't necessarily realize that they're not actually studying some categories. I don't think simply going by the order of the categories in Organize is enough - I think they need to specifically tell the app that these categories matter more and these ones matter less.
 

HW60

状元
I fully agree. Pleco could start every review session with an analysis of the actual flashcard database. The number of cards, number of overdue cards, average number of cards per review, expected number of due flashcards (based on the actual system without maximum number of flashcards) for the next days based on assumptions and more. If the number of overdue cards is small compared with the average number of cards per review, there is probably no problem at all, and using a system with a maximum number could only help if the expected flashcards for the next days are very unevenly distributed.

If there is a greater number of overdue cards however, Pleco could report the situation and make a proposal for the future SRS pool. The information "belonging to the SRS pool" is an additional information for every flashcard. I would remove all flashcards that are overdue more than a certain number of days - maybe 100 days - from the SRS pool and mark those flashcards as "not belonging to the SRS pool". In the next step I would remove all categories without cards in the SRS pool from the actual category selection. The result of these removals can be shown and explained to the user. For users with a one year review break the resulting category list would probably be empty.

Then a new category selection (the category SRS pool) is necessary. The user should decide which categories he wants to review and in which order. This is normally not the original order in organize cards, because it probably does not make much sense to repeat the first words from the first lessons. The user should assign a number to every category (or category group), and sorted by this number the categories can be automatically added to and removed from the category SRS pool. This sort order can theoretically be changed every day, though probably will stay somewhat stable over time. If the user selects a category without cards in the SRS pool, Pleco starts like in my example file from scratch with zero cards. Otherwise Pleco can put as many categories in the first review session as are needed to perform the maximum number of reviews chosen by the user, starting with category sort number 1 of the category selection. After the session the user can view statistical values like size of the SRS pool, change of the size, number of categories in the category SRS pool and change of this number, and many more.
 

daal

探花
Re your current problem, though, 'scoring' is actually stored at a profile level, not a scorefile level. (the thinking being that people might like to use different scoring systems in different profiles) Is there any chance that that might also explain why the tweak values had changed?

Yes, the problem was due to my getting lost in the interface. After tweaking the parameters for one profile, (Profile>Scoring>Tweak Parameters) I had been exiting Tweak Parameters and instead of exiting Scoring, I changed the scorefile (which had the same name as the profile, for example, the scorefile for my recognition profile is also named recognition) and then went back to Tweak Parameters thinking I was in a different profile. At least that's what I think was going on. So part of the confusion was caused by my naming scorefiles to match the profile, and part of it was caused by the interface not making it clear what I was tweaking.

At the moment, I am still finding myself often wanting to tweak and not knowing how to bring about the desired effect, for example, sometimes I start a test and only a few cards are in my queue, and it seems to me that whenever I start a test, the system should at the very least present all of the cards that I had gotten wrong in the previous test. I suspect that part of the reason is that some of my errant tweaks were still in place, but I'm not really sure, but in this context I'd like to suggest making it easy to reset such tweaks to their default settings.

Still looking forward to the new UI ::D
 

HW60

状元
If you are working on the flashcard system, I would like to make 2 minor proposals.
(a) In SRS, subject selection Alternating, it happens that Pinyin are displayed, asking for headword, but this is normally useless for one character words. I therefore added a profile for one character words. It would be more comfortable if Pleco would switch to e.g. show meaning ask for headword in these cases.
(b) 5 and more character words are not shown during review if Test type Fill-in-the-blanks is selected. The once in my life hint by Pleco probably nobody remembers. I therefore added a profile that I only use to remind me - then I change the Test type to Self-graded and repeat the same profile again. It would be much more comfortable if either Pleco would give the hint after every session or even would switch temporarily to Self-graded for every 5+ word.
 

mikelove

皇帝
Staff member
a) That should be fixable once we move to offering multiple-profiles-at-a-time testing - you'd simply have one profile for single character cards and one profile for multi-character cards. I'd rather stick to that approach than add another set of Show settings for single character cards - we've got too many settings already :)

b) Those are going to be shown during fill-in-the-blanks in our next major update.
 

HW60

状元
This is interesting, but my inclination would be to make removal a bit more obvious + flexible - provide a way to tell the system about your study priorities and then have the system respond by making it clear which priorities you're working towards now / which you're staying up on / which you're falling behind on / etc; if you only have time to study 100 cards today then you'll see a big angry red indicator that you're falling behind on priority X, and a few days or weeks after that you'll find that priority X has been disabled from active study entirely until you're ready to start up on it again.
Yesterday about 11 am my running watch told me: "so far you made 3000 steps. based on this you will probably make 9000 steps today. hurry up a little to reach your goal of 10000 steps today." (Garmin Insights). Maybe it is a good idea to let the Pleco user define his goals about categories and time to master them etc, and let Pleco estimate the user's success, e.g. "if you continue your flashcard reviews with 50 cards per day, then based on your previous results only 50% of your flashcards in these categories will have a score of 2000 after 100 days. If you want to reach your goal, you should increase your dayly reviews to 70."
BTW Garmin makes the goals for the users.
 
Top