I've posted this question in a couple of other forums, but haven't got any answers yet, so I thought I'd try my luck here.

I'm basically just curious to know how many of you have over 20,000 cards. How would or do you deal with using an SRS to learn, say Chinese, Japanese AND Korean if you wanted to do the “10,000 sentence” method for each? Wouldn’t the load eventually just become unbearable and take too long? Is there a limit or shortcoming to spaced repetition in this respect?

I’d appreciate any answers on this. I’m considering using Anki for Korean, but I intend to learn Chinese as well. The thought of spending several hours a day in the future doing SRS reps is rather frightening…

After about 7-10 reviews successful reviews, the cards expiration date is so long in the future, you'll be dead before it comes up again, so there is a top limit. As long as you keep the amount of new cards everyday below a certain threshhold, the review pile shouldn't grow too much.

Lol. That's true. If you guess a card right 10 times, the card is as good as deleted.

The exponential falloff in display frequency is what permits long-term use, as mentioned. If you have 20,000 cards all at one-year intervals distributed evenly across the year you're going to have about 55 cards/day on non-leap-years, which isn't unreasonable at all. You won't get even distribution, of course, but similarly your average interval will probably be a good bit more than one year by the time you hit 20,000 cards.

The only place where this breaks down is if you dump huge amounts of new cards into your SRS. Pace your additions (preferably with a built-in pacing mechanism; manually managing that sort of thing blows goats).

~J

woodwojr Wrote:preferably with a built-in pacing mechanism

Hmm, what exactly is this? Does Anki support it? Not sure what you mean, please elaborate. Thanks!

Anki does support it; it's their "new cards per day" setting.

What I mean is, if you dump a hundred cards into your SRS all in one day and it gives them all to you, you're going to do that hundred new cards, then they're all going to come due again in a few days, then they're all going to come due again in the next week, week and a half, and so on. If you add a hundred cards per day, soon you're being swamped, and even if you only do it a few times the spike can take a very long time to work itself out; until then, every so often you get a few days with an abnormally large number of reviews due.

The manual way to pace this is to just add a few cards a day, but that sucks. Some days you see a bunch of things you want to add, and if you're doing it manually you need to manually suspend and unsuspend the new cards, or even worse manually set aside the new material you want to add and remember to get back to it in the future.

So with the pacing mechanism, you set some maximum number of new cards to add per day; no matter how quickly you add material, you can't create a giant spike in due cards, because it gets presented to you (and thus scheduled for review) in a steady stream. Provided you pick the stream size well, the workload remains manageable.

(Probabilistically speaking, of course; the randomness in scheduling can create spikes, but their size is unlikely to be significant with large numbers of cards.)

Edit: timeboxing and itemboxing also solve this, but have the disadvantage of causing cards to be presented to you after they're due (as any items you don't get to are left sitting until someday you do get to them).

~J

I got just over 6000. With roughly 100 per day (but I am adding pretty much right now, a month ago it was 30/day). The deck is around 1 year old - so I can see myself in 5 or so years have 20000.
Anyhow - when you reasonably add stuff it should work.

You can dump an entire stack into Anki, you just don't have to review them all that day. Although I deleted cards from UBJG (personal reasons, it's a good resource), I now have about 2200 sentences (well, 4400 cards).

Like they say above, if you have spacing there's no problem. Anki has a default max spacing of 5 years. RevTK has a max spacing of 240 days (360 with easy button). Both programs try to spread out your reviews so you don't get "heavy bunches" days.

Reviews decrease overtime.

Just curious if this is true for other people, by saying you have 4400 sentences, you mean 4400 cards? Because each one of my cards has usually something like 2-5 sentences on it...

Squintox Wrote:Reviews decrease overtime.

It's actually the other way around, reviews won't ever decrease unless you lower the amount of new cards you enter. They will hit a limit however, if you limit your intake.

cloudstrife543 Wrote:Just curious if this is true for other people, by saying you have 4400 sentences, you mean 4400 cards? Because each one of my cards has usually something like 2-5 sentences on it...

That's a bad thing. Each card should be as short and focused as possible, that makes them way faster to review, much easier to grade and easier to remember. Having a long sentence from time to time to improve on grammar is a good thing, but several sentences per card? Nah, wouldn't recommend that, you're just making it harder for yourself. Split those sentences over several cards instead.

My girlfriend put anniversaries in my SRS. How can I pretend to forget now?

Quote:The exponential falloff in display frequency is what permits long-term use, as mentioned. If you have 20,000 cards all at one-year intervals distributed evenly across the year you're going to have about 55 cards/day on non-leap-years, which isn't unreasonable at all

It's not quite that simple because that's just the new cards. What about all the cards that have expired? If you always mark cards as hard in Anki (assuming you remember 100% of them), on day 1 you'd have 55 cards. On day 2, you'd have 110, and somewhere around day 5 you'd have 165 (day 1 cards due for the 2nd time + day 4 cards due for the 1st time as well as day 5 cards). And the less you remember the worse it gets

travis Wrote:

Quote:The exponential falloff in display frequency is what permits long-term use, as mentioned. If you have 20,000 cards all at one-year intervals distributed evenly across the year you're going to have about 55 cards/day on non-leap-years, which isn't unreasonable at all

It's not quite that simple because that's just the new cards. What about all the cards that have expired? If you always mark cards as hard in Anki (assuming you remember 100% of them), on day 1 you'd have 55 cards. On day 2, you'd have 110, and somewhere around day 5 you'd have 165 (day 1 cards due for the 2nd time + day 4 cards due for the 1st time as well as day 5 cards). And the less you remember the worse it gets

Yeah, but it doesn't matter. Say you enter 30 cards each day. That means you're reviewing 30 cards every day (counting the entering as a review). 3 days later, they start to expire, so you have 60 reviews each day (30 new ones, 30 expires). 7 days later or so, they expire again. Now you have 90 reviews each day. Add a few weeks and you get 120 reviews each day. Add a few MONTHS and you have 150 reviews each day. Add a YEAR or so and you have 180.

My point being, while the daily load will increase, the rate at which it increases loses momentum fast and will eventually hit a limit where you will have to wait 20 years or so before you get another 30+ to your daily reviews.

Of course, this is counting every card as passed, it isn't that simpel. But safe to say, if you only add 30 cards each day, you will never get 500 reviews a day, you will be dead long before that happens.

cloudstrife543 Wrote:Just curious if this is true for other people, by saying you have 4400 sentences, you mean 4400 cards? Because each one of my cards has usually something like 2-5 sentences on it...

I have 2200 sentences active, of which I do two types of cards so 4400 cards. A majority are single sentences mainly from iKnow. Some cards, mainly from Tae Kim, might have 2 or 3 sentences on them, especially the grammar cards. However, each cards have portions in bold lettering or set off to the side to point out that's the part of the sentence I'm concerned about. Makes it's easier to grade if I'm concerned with getting part of the sentence wrong.

Although I'd heed Tobberoth's advice about keeping things to one sentence if you can, it's not a hard rule. It depends on what you're doing the cards for. Recently, I've been thinking of making shadowing/reading cards, which will have upto a minute of dialogue on them. I want that as a card so it can be scheduled. Again, this is a specific case and not the norm.

Travis, if you're cards are building up like that (marking all of them hard, day after day), it's time to not add new cards and focus on stuff you're having trouble with. You should be wary of adding cards when there are things left to review. In addition, to don't have to review everything on the day it's due. Don't let the SRS dictate life to you. It's just a useful tool.

In the last few months, I've been timeboxing. So I've been waiting till I'm done with all the reviews before I add more items. If the review time takes long, it leaves less time to add new stuff (which I ALWAYS, ALWAYS review prior to adding). If the reviews don't take long, it leaves more time to study and add new stuff. Win, win.

Out of interest, anybody make a graph akin to Anki's study graphs on the following:

Adding XX number of cards per day, with a YY% pass rate on ALL reviews. Passed cards are spaced in a 3-4, 6-8, 12-16, 24-32, 48-64, etc block pattern. Failed cards are moved backward one step.

Based on the above, what will the graph look like with something like 50 cards at 80% or 30 cards at 90%, etc.

Yeah, I asked this question in December, and someone wrote a program in python that you can plug the data in to figure out how many cards per day you can add over time based on your average retention rate. It's in here somewhere.

EDIT: Here it is:
http://forum.koohii.com/showthread.php?tid=2245

Tobberoth Wrote:It's actually the other way around, reviews won't ever decrease unless you lower the amount of new cards you enter. They will hit a limit however, if you limit your intake.

Yeah that's what I meant, if you don't add anymore cards, your reviews decrease, and then if you start adding cards again, you can technically have the same amount of reviews when you 45,000 cards as 1,000 cards. That's the beauty of SRS.

bodhisamaya Wrote:My girlfriend put anniversaries in my SRS. How can I pretend to forget now?

Haha, fantastic :-)

travis Wrote:

Quote:The exponential falloff in display frequency is what permits long-term use, as mentioned. If you have 20,000 cards all at one-year intervals distributed evenly across the year you're going to have about 55 cards/day on non-leap-years, which isn't unreasonable at all

It's not quite that simple because that's just the new cards. What about all the cards that have expired?

No, that's no new cards at all. That's why I say one-year intervals; for simplicity of calculation, I'm assuming that this is a well-established deck with all mature cards. New cards are just added on top of this total, but because you can choose how many new cards to add you'll end up with a reasonable total (provided your idea of "reasonable total" is 56 or higher).

Failed cards will increase your load, but passed cards will decrease it substantially; assuming a 90% pass rate (low in my experience), you'll decrease your load much faster than you increase it.

Edit: not quite, actually. It's true that if your pass rate is that low, your workload for the current year will increase; you only see gains from increased scheduling length after the year is over. Still, this is with an exaggeratedly poor recall IMO; 97%-98% is more reasonable in my experience.

Quote:If you always mark cards as hard in Anki

Then you have other problems.

Quote:on day 1 you'd have 55 cards. On day 2, you'd have 110, and somewhere around day 5 you'd have 165 (day 1 cards due for the 2nd time + day 4 cards due for the 1st time as well as day 5 cards). And the less you remember the worse it gets

I don't believe this is true. As far as I can tell, Hard reschedules the card at the same interval it was at before, so modulo the random jitter added in, you'd have 55 cards/day off into eternity.

~J

... which is why you should read the old thread, download and run the python script, and run it to see what *actually* happens, rather than just speculating.

You can plug in your pass rate for mature cards, and the number of new cards you add per day,and it will tell you the number of reviews per day you'll wind up doing over a year. Just scroll down halfway down the thread, and look for iSoron's posts. He wrote the script.

I saw the script; the description suggested that it wouldn't be much better than mental estimation.

~J

woodwojr Wrote:

Quote:If you always mark cards as hard in Anki

Then you have other problems.

Agreed, this is a bad thing to do.

Every time you mark a card as hard, Anki assumes the card *is* hard, and decreases the rate at which its interval grows. After answering 'hard' for a card several times, the card's rate will reach the bottom, and its interval will almost stop expanding. If you do this for every card, and you keep adding new cards at a constant rate, your reviews will grow almost linearly over time.

woodwojr Wrote:I saw the script; the description suggested that it wouldn't be much better than mental estimation.

But only if you happen to use the same parameters I've used.
Changing them even slightly can cause wide variations.

That graph's shows in part why I changed up my habits. I give myself a set amount of time to study, and not a set amount of material. If there's to much to review in the time you dictated, hold the rest off till the next day. If there's not enough material to review (really true at the beginning with an empty deck), then add more stuff till your time is out.

That's also why I set Anki to add new cards at the end of reviews.

In this way, I never look at is as: Oh, I have 800 cards to study. I look at is as: Today is the same as yesterday as it will be tomorrow. Two hours to study (6 time boxes of 20 minutes each).

iSoron Wrote:Every time you mark a card as hard, Anki assumes the card *is* hard, and decreases the rate at which its interval grows. After answering 'hard' for a card several times, the card's rate will reach the bottom, and its interval will almost stop expanding. If you do this for every card, and you keep adding new cards at a constant rate, your reviews will grow almost linearly over time.

Hey I didn't know that! Does that mean that it's better ーspeaking in terms of projected review timeー to actually mark the cards as failed instead of hard?