Here's a question that's been kind of bugging me. I've fallen into a pattern that kind of goes like this:

1. I'll add a chunk of new cards over time. Say, a month to 6 weeks or so. I'll get into a good groove, and things are going well.
2. Reviews per day start to increase, and before I know it, I get swamped in reviews. The number of reviews I have to do per day shoots up to something like 2-300, and RL obligations are still abundant. Sometimes I'll have as many as 5-600 in the pile. (Ugh.)
3. So I wind up not adding new cards until I catch up, and the number of cards to review per day drops back down to something like 100 or so. Sometimes I'll go 2-3 weeks without adding any new cards in these phases.
4. So I start to get frustrated at my lack of progress and start adding new cards in large chunks again. Go back to step 1.

Other than just not adding so many cards at a time, is there a way to make progress and not get swamped? Because I don't see it.

Right now I'm doing okay, but I'm not making nearly as much progress as I'd like to.

If I just keep adding massive amounts of vocab, will I be able to just power through and speed up my review times to the point where doing an extra 100 reviews a day won't matter? Or will I wind up courting burnout?

Does Khatzu really? My philosophy is to just read a lot. I'm totally hooked on manga right now and I really can't help but add around 30 a day. It's just habit for me to add stuff, I don't even think about the consequences. But then again, I've been getting around 230 reviews a day lately. But then again, it's kinda fun. Most of the sentences in my deck are all really cool sentences that bring me back to the story and context I got them from. It's like I get to revisit the cool parts of the story I got the sentences from. (This will sound totally wussy, but there are 2 or 3 cards in my deck that actually make me cry when I read them - literally. It was a really depressing story I got those from about a poor dog who had to deal with a lot of bad people.)

Once I'm at the point where I've added 30 new cards in an day, I realize how it may hurt the next day. Well, not hurt, but distract from more actual reading time. So at that point I stop and do reviews. And then watch something instead of reading, since I don't get nearly as many sentences out of the stuff I hear. Not uncommon... unless there are Japanese subtitles?

I used to have this problem too. I decided to hard-limit myself to 100 cards a day, where 1 "card" can mean either 1 review or 1 addition. It sure made SRS slow, but by then I was already pretty good at Japanese and I was finding SRS was actually taking away time I could be *USING* the language. (Example: I put off playing FF7 in Japanese because "My SRS can't accomodate that many sentences right now")

Found myself doing what snispilbor mentioned, which I found unacceptable, so I 1: ditched all my production cards and 2: limited SRS entry. I don't really enter many. I enter in sets of 15, and I'll either enter 1 or 2 sets in a day. I keep a text file of all the sentences I've pulled out of media that I haven't entered yet, and as my Japanese activity waxes and wanes, the backlog builds up and winds back down over time. Although it's best not to let the backlog get too great, because then by the time you go to enter a sentence, you don't remember it's context at all anymore :o

It's probably possible to just power through all of it though. But, I think the risk of burnout is too great, at least for me personally. I know my limitations, and burning out is a real risk for me, so I need to manage myself and what I'm doing to make sure that I don't overload.

The thing with an SRS is, when you add a card, it's added. It won't stop showing up. Which means, no matter how few cards you add every day, you are constantly building it up, reviews will go up no matter what. If you add 50 sentences a day, you'll get 300 reviews quickly, if you add 1 each day it will take years, but it will happen and there's really no way to get around it. This is the problem with SRS and the only solution is to have some system like... when you have reviewed a sentence 20 times, it's automatically removed. Otherwise, that 21st review of that card will hit at the same time as a 20th review card, a 19th review card etc, ad infinitum.

Tobberoth Wrote:This is the problem with SRS and the only solution is to have some system like... when you have reviewed a sentence 20 times, it's automatically removed.

The ever increasing delay already takes care of that. For example, if the initial delay is 1 day and you double it for each successful answer, after 20 correct answers the delay would be 2^20 = 1'048'576 days or ~ 2'870 years.

So 20 times is a bit extreme but even with more reasonable values, there is going to be a point when it's likely you'll simply stop using that SRS before the cards are up for review again.

Now this site's SRS has a maximum interval cap, so you'll keep reviewing forever unless you decide to stop at some point.

Here's a script that simulates a simplified SRS. Variables are pass rates, failure and success multipliers, initial card interval and, of course, number of new cards per day. It does not take easiness into account, and it may contain a few bugs.

With these settings:
- Pass rates: 70% (young cards) and 85% (mature cards)
- Failure multiplier: 0.5
- Success multiplier: 2.0
- Initial card interval: 3 days

you get these graphs:
- 15 cards per day
- 30 cards per day
- 50 cards per day
- 100 cards per day

Long story short, if you keep adding x cards/day (with these multipliers, initial intervals and pass rates), you can expect your average workload to stabilize, after a year, somewhere around 13x cards.

Edit: Young/mature cards pass rates.

Hmm... I tried running it, and I got syntax errors. Then again, I know nothing about Python. I'm more of a 6502 Assembly Language and BASIC (when it had line numbers) guy. >_>a

EDIT: It winds up highlighting the "i" right after:

print i, count, avg

near the very end.

I'm just running python 3.0 off of the website. Maybe they changed something?

Wow, even with as few as 15 new cards a day you still hit approximately 200 reviews a day... Then again, it depends on failure rates... my rate for sentences seem to be close to 90% when fresh, 99% when mature. I don't know how much of a difference that would make.

Tobberoth Wrote:Wow, even with as few as 15 new cards a day you still hit approximately 200 reviews a day... Then again, it depends on failure rates... my rate for sentences seem to be close to 90% when fresh, 99% when mature. I don't know how much of a difference that would make.

I'm at 1900 now, I've been adding an average of 18 per day... My reviews have only once gone over 100.

My rates are about 80%-90%.

My Heisig reviews stationed at ~300 for a long time.

The only time It got above 100 was when I mass imported cards from other decks, but it stayed there for only a short time.

Besides that, I was adding ~30 cards/day, but recently I'm slowing down. My sentences review is never above 60. My due cards graph in anki drops abruptly after 1 week.

I'm with ~2500 sentence cards right now, plus 2042 from RTK.

I'm at about 6,000 cards, with about 3,000 facts, with my retention on young cards at 89%, and 96.8% on mature cards. Then again, I've been at this deck for 1.3 years, so I have a LOT of cards that are spaced out far into the future, so my average daily load these days is under 100 cards a day, but that's because I haven't been adding cards much in the last 3 weeks or so.

Hey, is there any way to create a simple web app where we could just dump in our numbers and get a rough idea of how many cards we can safely add without drowning?

rich_f Wrote:Hmm... I tried running it, and I got syntax errors. Then again, I know nothing about Python. I'm more of a 6502 Assembly Language and BASIC (when it had line numbers) guy. >_>a

EDIT: It winds up highlighting the "i" right after:

print i, count, avg

near the very end.

I'm just running python 3.0 off of the website. Maybe they changed something?

Yes, you need Python 2.x -- 2.5 should work. Or try running the script through 2to3, which should have been included in your 3.0 installation (http://docs.python.org/dev/3.0/library/2to3.html)

Python 3.0 has breaking changes vis-a-vis 2.x, and one of those is that print is now a function. So the third thing you can do is try putting parentheses around all the arguments to print in the script, and that might fix it.

iSoron Wrote:if you keep adding x cards/day (with these multipliers, initial intervals and pass rates), you can expect your average workload to stabilize, after a year, somewhere around 13x cards.

Actually, under the conditions you describe, it will never reach a limit. I already did the maths and found that the curve will continue increasing logarithmically. Your graphs confirm that.

Of course logarithmic growth is not so bad. The rate of growth slows down over time, and after enough years it will appear to be constant because you won't live long enough to see much increase.

However, another way of looking at it is that you only do a fixed number of reviews a day. You revise old questions first, and if and only if you have vacant slots, you draw on new material.

Under those conditions, it means that the number of new questions you add will slowly decrease over time. In other words, you have a fixed amount of time a day to study (you can't spend more than 24 hours), so if you want to keep revising everything you've learned albeit at exponentially increasing intervals, then the rate of new material must slow down over time.

I'd be curious to see some simulations of the fixed number of reviews per day method.

Raichu Wrote:However, another way of looking at it is that you only do a fixed number of reviews a day. You revise old questions first, and if and only if you have vacant slots, you draw on new material.

Script updated. You can now set the number "slots" per day. Note that I assume you always review every expired card, regardless of this limit, but you only draw new cards if your review count is below the number of slots.

Here are some graphs showing the total number of cards over time.
Max of 30 new cards per day. Same settings as previous graphs.

- 100 slots
- 200 slots
- 300 slots

The rate of growth does not seem to slow much, after all.
I'm starting to like this SRSing strategy.

Note: For those of you wondering, the slots limit does seems to be effective - the average number of expired cards per day does not increase (much?) over time. Here is a graph showing the number of expired cards (30 max new cards/day and 200 slots) over a period of 10 years.

Raichu Wrote:However, another way of looking at it is that you only do a fixed number of reviews a day. You revise old questions first, and if and only if you have vacant slots, you draw on new material.

I wish Anki had an option like that - rather than set the Number of new cards a day, at say, 30, I would set a fixed number of reviews to 100...

I'm pretty sure you could manage that with Anki. It has options to show the due cards in several orders... You can also tell it when you want the new cards to pop up[beginning, randomly dispersed, or end].

Interesting stuff. I finally got the script to run under 2.6.1. With my retention rate for long-term cards at 96.8%, I noticed that the curve flattens out after a certain point. With 30 cards a day, by day 299, it hovers around 220 cards a day, which is what it was 100 days prior (almost).

But with 15 cards a day, it hovers around 110 cards a day.

So I guess my answer is that if I can find a way to hand 220 cards a day, if I add 30 a day, that means I could handle 9000 new cards in 300 days. But if I limit it to 15, I can easily handle 110 a day, for 4500. Maybe there's a happy medium in between?

I guess the answer is going to be totally different for everyone based on his/her retention rate, huh?

rich_f Wrote:I guess the answer is going to be totally different for everyone based on his/her retention rate, huh?

Yeah, I think so.

Other very important variables are the multipliers. RevTK's SRS always use 2 (roughly) for success multiplier and 0 for failure multiplier. Anki lets you specify the failure multiplier (default is 0.1, I think), and sets different values for success multipliers for each card based on your previous answers. It's called 'Ease' on the card statistics and 'factor' on the database.

The higher the success multiplier, the faster the curve flattens, so if you want accurate results, make sure you set these variables appropriately too.

rich_f, do you use the feature of anki which tells it to only show you a certain number of new cards each day? I find that very helpful, because I can just keep adding new cards in whenever I want, and they just build up in a queue instead of swamping me with work all at once. Another good side effect is that when I finally see the card come up, I might decide that I don't really need that card after all, and delete it.

iSoron Wrote:Here's a script that simulates a simplified SRS. Variables are pass rates, failure and success multipliers, initial card interval and, of course, number of new cards per day. It does not take easiness into account, and it may contain a few bugs.

With these settings:
- Pass rates: 70% (young cards) and 85% (mature cards)
- Failure multiplier: 0.5
- Success multiplier: 2.0
- Initial card interval: 3 days

you get these graphs:
- 15 cards per day
- 30 cards per day
- 50 cards per day
- 100 cards per day

Long story short, if you keep adding x cards/day (with these multipliers, initial intervals and pass rates), you can expect your average workload to stabilize, after a year, somewhere around 13x cards.

Edit: Young/mature cards pass rates.

Interesting. I don't find that the number of reviews goes up at such a high rate, but I guess its because I normally get better retention rates than the inputs you used.

I think the simulator is different from reality in the way that failures are distributed. It gives equal failure rates to all cards of a certain interval, regardless of ease or number of times reviewed. I don't know how to explain it concisely but I think that has a big effect on the results of the simulation.

Note that in Anki there are independent "interval" and "factor" variables.

On pass, due = interva; interval = interval * factor;
On fail, due = 0; interval = start_interval; factor -= 0.2; clamp factor above 1.3;

Zarxrax Wrote:rich_f, do you use the feature of anki which tells it to only show you a certain number of new cards each day?

Nope. The reason why is simple-- if Anki says a card is due, then I want to make sure I study that card as soon as possible. If I put an artificial limit on the number of cards I can see in a day, then I'll be short-changing myself either on the short-term or long-term end. (You have to cut it off somewhere.)

That's part of the reason why I was very much in favor of having multiple review models in Anki, because I disagreed with the new model that Damien picked for 0.9.9.0 (I think that was the release), where it pushed old cards up front at the expense of younger cards. The way the program is now, with the user's choice of 4 separate review models, is perfect for me.

My only question was really the one iSoron solved for me with his Python script. Dude, that was awesome. Thanks again for that.

Now it's just a question of implementation on my part.

Rich_f, I think he was refering to completely new cards which hasn't been reviewed yet, not expired cards.

Oh duh. Sorry about that. I don't add so many cards that it's an issue, really. My way of adding new cards is no limits on new cards (not that I need it), I sprinkle them in with due cards, and in random order, just to torture myself. That way there's less chance of me associating new cards with each other, and less chance of not seeing them if I don't clear my pile every day.