Back

Suggestions for maximum review time on Koohii SRS

#1
As an off-and-on user of this site for over a decade, I am naturally very interested in long-term maintenance of knowledge. If I understand theReview Scheduling rules, reviews are never scheduled for over 240 days. Is that correct? Is there a way for a longer (or unlimited) maximum delay? I usually go “off again” from the daily grind of reviewing double digits of 大, 日, 二, etc. every day even though I’m never going to forget them, but without ongoing maintenance, I soon forget brain teasers like 襲... 

Thanks for a great site!
Edited: 2017-10-27, 8:04 am by ファブリス
Reply
#2
(2017-10-26, 5:46 pm)mspertus Wrote: Is there a way for a longer (or unlimited) maximum delay?

Anki. I have hundreds of cards now with intervals over 10 years. My cards that are the furthest out won't come up for review until 2029. Cool
Reply
#3
@mspertus

On the subject of Kanji Koohii then...

Technically it's close to a year if you answer "Easy" for cards in box 8 (7+ reviews), since EAsy answer = 1.5x interval (plus or minus some variance which helps spread due cards).

mspertus Wrote:I usually go “off again” from the daily grind of reviewing double digits of 大, 日, 二, etc. every day even though (...)

I don't understand what you mean here. The super easy kanji whether you answer yes or east, will have at least 240 days interval after 7 successive reviews. And close to a year if you answered "Easy".

Regarding upcoming changes to the maximum delay on Koohii SRS

I was talking to someone recently about that, and we agreed on a new default for Kanji Koohii SRS that increases the maximum review interval to ~approx. 5 years (~1900 days).

3 7 13 26 53 109 223 457 936 1919

So for the first 7 reviews (in bold) the intervals are a little bit less than the current values (multiplier of approx. 2.05 ).

The boxes will be increased to 10. So after 10 reviews you have an interval of ~5.2 years, and over 7.5 years if you answer "Easy" for those.

I confirm this is the next update. So this is a good time to make suggestions for those settings.
Reply
Thanksgiving Sale: 30% OFF Basic, Premium & Premium PLUS Subscriptions! (Nov 13 - 22)
JapanesePod101
#4
What will be the impact, if any, on cards that are currently in the boxes that are having their intervals changed? I assume the change will only have an effect when cards are re-scheduled for their next review, but want to ensure we don't all of a sudden have large number of cards coming due because of the shorter intervals.

Will the cards wind up being "out of order" because of this? For example, one day, box 5 cards will have a 60 day delay (on average), then the change goes into effect, now the next set of box 5 cards will have a 53 day delay, and will come due before the ones that were passed the day before.

Not this is a big deal or anything (the ordering/clustering is not really a desired feature anyway), just trying to understand what to expect after the change.
Reply
#5
No, this change should not affect any due cards at all. The card's due time is calculated at the moment you review them (precisely, when they are synced with the server Wink). Only as you review due &new cards, the new intervals will be calculated based on new defaults (including Hard answer), AND the cards in box 8 (7+ reviews) will move to box 9, 10 and 11 (as seen in the flashcards list page).
Edited: 2017-10-27, 3:13 pm
Reply
#6
(2017-10-27, 8:03 am)ファブリス Wrote: Regarding  upcoming changes to the maximum delay on Koohii SRS

I was talking to someone recently about that, and we agreed on a new default for Kanji Koohii SRS that increases the maximum review interval to ~approx. 5 years (~1900 days).
Since the word default is flying around, are there any plans to add custom settings for the SRS? Like setting maximum review interval or review multiplier.

Unless they're already there and I just can't find them, in which case I wouldn't mind being pointed in the right direction. Smile
Reply
#7
There are no plans at this time, no.

What custom multiplier would you use?

And how would a maximum interval work? On Koohii that would mean increasing the maximum Leitner box (each box having a greater interval). Five to seven years of interval is not enough?
Reply
#8
Te upcoming changes sounds fantastic! Pretty much just what I was hoping for. FWIW, the confusing "double digit" comment meant that if you have thousands of cards at 240 day delay, more than 10 will be due every day. Finally, as requested, I do have one suggestion. If someone enters "hard," the next review should never be more than 240 days. If I say something is Hard, I probably don't want to wait years to see it.

(2017-10-27, 8:03 am)ファブリス Wrote: @mspertus

On the subject of Kanji Koohii then...

Technically  it's close to a year if you answer "Easy" for cards in box 8 (7+ reviews), since EAsy answer = 1.5x interval (plus or minus some variance which helps spread due cards).

mspertus Wrote:I usually go “off again” from the daily grind of reviewing double digits of 大, 日, 二, etc. every day even though (...)

I don't understand what you mean here. The super easy kanji whether you answer yes or east, will have at least 240 days interval after 7 successive reviews. And close to a year if you answered "Easy".

Regarding  upcoming changes to the maximum delay on Koohii SRS

I was talking to someone recently about that, and we agreed on a new default for Kanji Koohii SRS that increases the maximum review interval to ~approx. 5 years (~1900 days).

3 7 13 26 53 109 223 457 936 1919

So for the first 7 reviews (in bold) the intervals are a little bit less than the current values (multiplier of approx. 2.05 ).

The boxes will be increased to 10. So after 10 reviews you have an interval of ~5.2 years, and over 7.5 years if you answer "Easy" for those.

I confirm  this is the next update. So this is a good time to make suggestions for those settings.
Reply
#9
Regarding Hard answer that's a very good point, which I didn't think about. Indeed this becomes an issue when adding more boxes and very long intervals.

Currently the "Hard" answer demotes the card, and then reschedules according to the lower box interval. We could clamp this down to something even lower, eg. box4 , or box 5 along with its interval.
Reply
#10
(2017-10-28, 10:11 am)ファブリス Wrote: There are no plans at this time, no.

What custom multiplier would you use?

And how would a maximum interval work? On Koohii that would mean increasing the maximum Leitner box (each box having a greater interval). Five to seven years of interval is not enough?
My bad since I didn't specify it.

When it comes to maximum interval shortening it was more what I had in mind. In case of Koohii that could probably work as cutting off review boxes, so I could just use 7 out of 10 review boxes for example.

You could further customize that with custom multiplier, setting different interval between boxes, but being able to limit the maximum interval is mostly what I'm after.
Reply
#11
(2017-10-29, 7:44 am)rost400 Wrote: When it comes to maximum interval shortening it was more what I had in mind. In case of Koohii that could probably work as cutting off review boxes, so I could just use 7 out of 10 review boxes for example.

You could further customize that with custom multiplier, setting different interval between boxes, but being able to limit the maximum interval is mostly what I'm after.

I, too, think that the maximum interval that has been decided on is way too long.  Seriously, five years?  And not for the sake of improving learning but for the sake of reducing the workload for those who put off their daily reviews?  If you work through RTK at a sane pace and do your daily reviews consistently, the workload never even gets close to being overwhelming.

I have been having trouble with the kanji in box 7 as it is; by the time they come around, unless they are kanji I knew long before starting on RTK, I have no recollection of ever having seen them.  In fact, the trouble happens somewhere in between.  One of the middle intervals is too long.  I find that I remember kanji very well up to the fourth and fifth repetition, and then I fail on the next one.  Then, the next time around, the same thing happens again.  The worst consequence of this is that when I see certain kanji on TV or in print, I can't recognize them, despite having studied them on this site.  Shortening one of the earlier intervals would, I suspect, greatly help with recalling the kanji.

Before Fabrice claims that the problem is with the way I'm studying RTK, let me point out that I almost always make up my own stories, which tend to be very vivid.  When a story doesn't stick, I change it.  When I can't recall a kanji, I click "No".  I never use "Easy".  I click "Hard" if I took a long time to remember a kanji. The problem I described above is not a problem with me; it's a problem with the way the brain saves information.  If you haven't seen something in a long time, your brain decides it's not essential and discards it to make space for something it considers more important.

The best solution would be to allow the user to change the default based on what works best for the particular user.  But I won't hold my breath, based on the response to my last request.  In the end, this change will make it harder for me to continue with RTK, even though I have only 156 kanji left to complete RTK1.  What's the point of spending hours making up elaborate stories if you won't be able to remember the kanji in the end?
Reply
#12
In fact, the original poster's request could have been addressed much better by changing the way Easy and Hard work.  Instead of increasing the interval by a factor of 1.5, "Easy" could simply send the kanji to the last bin. 

Note that the original poster's concern was with easy kanji making the reviews boring and discouraging users from sticking with RTK.  I must say I don't understand this.  How hard is it to write down the easy kanji and push "Easy"?  Over time, one sees the easy kanji less and less often.  In any case, if a user is so bothered by kanji like 大, 日, and 二, they can simply delete them so they are not included in any reviews.  There is no need to ruin this site for everyone else.
Edited: 2017-10-30, 11:19 am
Reply
#13
(2017-10-30, 11:18 am)mercury Wrote: In fact, the original poster's request could have been addressed much better by changing the way Easy and Hard work.  Instead of increasing the interval by a factor of 1.5, "Easy" could simply send the kanji to the last bin. 

Note that the original poster's concern was with easy kanji making the reviews boring and discouraging users from sticking with RTK.  I must say I don't understand this.  How hard is it to write down the easy kanji and push "Easy"?  Over time, one sees the easy kanji less and less often.  In any case, if a user is so bothered by kanji like 大, 日, and  二, they can simply delete them so they are not included in any reviews.  There is no need to ruin this site for everyone else.

OP here. My concern was with easy kanji making the reviews boring *after RTK was completed,* so having easy move things to the last bucket doesn't really help. More specifically, I want it to be possible to go in "maintenance mode" where one can do reviews a couple of times a week without having to keep cycling through the 1500 kanji I'll never forget. Without an efficient way to maintain knowledge of the kanji, I invariably stopped using RTK once I completed it. Fabrice's plan seems like a good approach to me.
Reply
#14
I'll look at adding settings : max box & multiplier (intervals) mainly. This should answer needs of users both advancing through RTK and users reviewing after completing RTK.
Reply
#15
(2017-10-30, 1:34 pm)mspertus Wrote: OP here. My concern was with easy kanji making the reviews boring *after RTK was completed,* so having easy move things to the last bucket doesn't really help. More specifically, I want it to be possible to go in "maintenance mode" where one can do reviews a couple of times a week without having to keep cycling through the 1500 kanji I'll never forget. Without an efficient way to maintain knowledge of the kanji, I invariably stopped using RTK once I completed it. Fabrice's plan seems like a good approach to me.

If you will never forget those 1500 kanji, why not just delete them so they are never included in your reviews?  Deleting a kanji just removes it from your list; it doesn't delete the kanji page or the story you made up for it.  You can always add a kanji back if you find that you are, after all, forgetting it. 



(2017-10-30, 1:52 pm)ファブリス Wrote: I'll look at adding settings : max box & multiplier (intervals) mainly. This should answer needs of users both advancing through RTK and users reviewing after completing RTK.

Wow; thank you!  It would be great to be able to reduce the maximum to, say, the sixth or even the fifth or fourth box, changing it back if and when one finds one is remembering the kanji better.  The ability to change the multiplier could allow for further experimentation.  If this is presented as an "Advanced" setting, it won't confuse new members who have not used the system long enough to know that they prefer something different from the default. 

It took me a while to learn to use Kanji Koohii as it is intended.  Lately, my only issue with it has been the length of the intervals.  In my opinion, this  kind of flexibility would make the site even better.
Edited: 2017-10-30, 3:20 pm
Reply
#16
(2017-10-30, 3:04 pm)mercury Wrote: If you will never forget those 1500 kanji, why not just delete them so they are never included in your reviews? 

Having ever increasing intervals is a much more elegant solution to the problem. The whole point of an electronic SRS system is so that you don't have to do this kind of maintenance manually.
Reply
#17
(2017-10-30, 1:52 pm)ファブリス Wrote: I'll look at adding settings : max box & multiplier (intervals) mainly. This should answer needs of users both advancing through RTK and users reviewing after completing RTK.

Thank you for considering this, much appreciated. Smile
Reply
#18
(2017-10-30, 1:52 pm)ファブリス Wrote: I'll look at adding settings : max box & multiplier (intervals) mainly. This should answer needs of users both advancing through RTK and users reviewing after completing RTK.

sounds like a good solution for everyone to get what they want. 
For me, I thought your original idea of adding boxes brilliant! I feel that kanji that I encounter anyway (in reading books mostly) are sometimes “too easy ” i.e.  could have really long intervals, while others are muddier and might even need to be restudied and go back through boxes 1-7 (-10, hopefully ;-) ) again. The adding of boxes (together with use of easy and hard buttons) sounds like a great way to have more time/energy for the "non-stick" cards.
Edited: 2017-11-03, 12:36 pm
Reply
#19
Sorry it's taking time, Life stuffs.

(2017-10-28, 10:33 am)mspertus Wrote: (...)  If someone enters "hard," the next review should never be more than 240 days. If I say something is Hard, I probably don't want to wait years to see it.

Re: Hard answer

The straightforward implmentation for me is to add a "Max Box for Hard Answer". So you'd want to pick a box with approx 240 ish days.

If you pick box 4, and it's 30 days, then that means a Hard answer to a card in box 10 would drop all the way down to box 4, with box 4 interval.

If I add a setting in days, instead.. then I need to match that to a box somehow otherwise the SRS graph is less meaningfull. I don't think it's a good idea to have a card drop from box 10 to box 9, but sopmeow get a 30 day interval. Over time the graph would be leass meaningful. With Leitner we want the intervals to more or less match the box, so they represent not just positive review streaks, but also long term retention.

So, instead of trying to match a custom Hard answer interval to a box, I think I'll just add a "Max Box for Hard Answer".
Reply
#20
It sounds like this point has already been made and will probably be answered, but on the assumption that every vote counts (and having been absent when my countrymen elected a reality show host as president).... 

I would like the keep using the site (which I LOVE, by the way) as it is now.  I have been using it for many years, and I am trying to pass the 漢字検定準一級.  I can't afford to get many wrong. 

I'm happy to review cards every day, and I require myself to get all pronunciations of the character to give myself a pass.  In other words, I'm one who is not necessarily using the site in the recommended way, but I am definitely a loyal user / member. 

If the change would present a choice for "no change" or limited change, that would be great.
Reply
#21
(2017-10-26, 5:46 pm)mspertus Wrote: As an off-and-on user of this site for over a decade, but without ongoing maintenance, I soon forget brain teasers like 襲... 

SRS is an awesome tool but to worry about individual kanji after 10 years? SRS is meant to  bootstrap language learning until we can handle native materials. At that point native materials become our natural 'SRS'. At this point i think native materials would be much more helpful remembering kanji. Take the leap mspertus. I believe in you! ;D
Reply
#22
OK, one last appeal.  It appears some of us would be VERY happy if we had the option to keep seven boxes and not go to 10 -- or, alternatively, some way to shorten the maximum interval further.  Five years will be too far for some / many of us. 

I would be remiss if I did not add every time I wrote that I love the site and am deeply appreciative of everything you have done and are doing to maintain it.
Reply
#23
I wrote a little Perl script to experiment with various settings, to get a feel for what affect tweaking would have (code below):

$ ./rtk-knobs.pl 10 4 3 2.05 0.15 1.5  #(note, these are the script's default values)


Adjustable Knobs
  N = 10   number of boxes
  H = 4    maximum box for cards marked 'Hard'
  F = 3    interval for first box
  M = 2.05 interval factor
  V = 0.15 variance factor
  E = 1.5  easy factor

Interval assigned to card when placed in box N
(values are rounded up after all calculations are done)
  Nominal value = F * (M ** (N-1))
  Nominal range = Nominal +/- (V * Nominal)
  Easy columns = E * Nominal
  'Hard' box is indicated, assuming hard cards take on box interval

  box  1:     3 (    3 ..     4)  ||      5 (    4 ..     6)
  box  2:     7 (    6 ..     8)  ||     10 (    8 ..    11)
  box  3:    13 (   11 ..    15)  ||     19 (   17 ..    22)
 HARD  4:    26 (   22 ..    30)  ||     39 (   33 ..    45)
  box  5:    53 (   46 ..    61)  ||     80 (   68 ..    92)
  box  6:   109 (   93 ..   125)  ||    163 (  139 ..   188)
  box  7:   223 (  190 ..   257)  ||    334 (  284 ..   385)
  box  8:   457 (  388 ..   525)  ||    685 (  582 ..   788)
  box  9:   936 (  796 ..  1077)  ||   1404 ( 1194 ..  1615)
  box 10:  1919 ( 1631 ..  2206)  ||   2878 ( 2446 ..  3309)

Here's the code if you want to experiment with different knob settings ....

Code:
#!/usr/bin/perl

use strict;
use POSIX;

our $boxes = shift || 10;
our $hard  = shift ||  4;
our $first = shift ||  3.00;
our $mult  = shift ||  2.05;
our $fuzz  = shift ||  0.15;
our $ease  = shift ||  1.50;

sub nth {
 my $n = shift;
 my $e = shift || 1.0;

   my $b = $first * ($mult ** ($n - 1));
   my $f = $b * $fuzz;

   my $min = POSIX::ceil($e * ($b - $f) );
   my $max = POSIX::ceil($e * ($b + $f) );
   my $val = POSIX::ceil($e * $b);

   return ($min, $val, $max);
}

print "\n";
print "Adjustable Knobs\n";
print "  N = $boxes \tnumber of boxes\n";
print "  H = $hard  \tmaximum box for cards marked 'Hard'\n";
print "  F = $first \tinterval for first box\n";
print "  M = $mult \tinterval factor\n";
print "  V = $fuzz \tvariance factor\n";
print "  E = $ease \teasy factor\n";
print "\n";
print "Interval assigned to card when placed in box N\n";
print "(values are rounded up after all calculations are done)\n";
print "  Nominal value = F * (M ** (N-1))\n";
print "  Nominal range = Nominal +/- (V * Nominal)\n";
print "  Easy columns = E * Nominal\n";
print "  'Hard' box is indicated, assuming hard cards take on box interval\n";
print "\n";

for my $i (1 .. $boxes) {
   my ($l,  $m,  $h)  = nth($i);
   my ($el, $em, $eh) = nth($i, $ease);
   printf " %4s %2d: %5d (%5d .. %5d)  ||  %5d (%5d .. %5d)\n",
       ($hard == $i)?"HARD" : " box",
       $i,
       $m,  $l,  $h,
       $em, $el, $eh
   ;
}

__END__
Edited: 2017-11-09, 6:54 pm
Reply
#24
(2017-11-06, 8:26 pm)juniperpansy Wrote:
(2017-10-26, 5:46 pm)mspertus Wrote: As an off-and-on user of this site for over a decade, but without ongoing maintenance, I soon forget brain teasers like 襲... 

SRS is an awesome tool but to worry about individual kanji after 10 years? SRS is meant to  bootstrap language learning until we can handle native materials. At that point native materials become our natural 'SRS'. At this point i think native materials would be much more helpful remembering kanji. Take the leap mspertus. I believe in you! ;D
I do read some native materials (mainly easy-ish manga) and struggled my way through Japan mainly without using English when I visited a few months ago. However, with two jobs and until recently a child at home, I often have to put serious study aside for long periods of time (and even in the best of circumstances my time is very limited). If I could reinforce my knowledge more efficiently, I would still have all the kanji and keywords at my fingertips when I come back to serious usage (I do something similar with vocabulary in anki), so I am very excited about the upcoming changes.

That being said, I understand that not everyone's circumstances or affinities are like mine, so I also like the knobs.
Edited: 2017-11-09, 11:28 pm
Reply
#25
Here is what I've got so far. Let me know if I can make the text clearer.

The intervals update when you select different values.

I'd prefer to put sensible choicesi n the multiplier instead of validating possibly nonsense multipliers. What would you like to see there in terms of smallest realistic multiplier? Or Greatest?

[Image: UuVWyxe.png]
Reply