[Bug] The Random option in the Review sort order setting is not truly random

[vaibhav]

If related cards are due on the same day, Anki tends to show them one after the other during review or after 2-3 other cards.

By related cards, I don’t mean the sibling cards (which are buried anyway). I am talking about the cards that are related to the same topic, were created on the same day and were first reviewed on the same (or consecutive) days.

This happens despite the fact that I have set the Review Sort Order setting in the Deck Options to “Due date, then random”. (And it is not the case that these cards are some of the very few cards due on a particular day.)

If I create a filtered deck with the due cards and then choose “Random” in the “cards selected by” option, the problem disappears, and these cards are no longer shown close to each other.

So, it makes me think that the random option in the Review Sort Order setting is not truly random, while that in the “cards selected by” option in filtered decks is.

I am doing my reviews on AnkiDroid v2.16alpha97 with the new backend and v3 scheduler enabled. So, it should behave similarly to Anki 2.1.61. However, I not tested this with Desktop Anki because I do most of my reviews when I am not at my computer.

Also, I want to request for an option in the Review Sort Order setting to sort reviews randomly without considering the due date at all. The reason is that if Anki also considers the due date, it first shows all the cards that were buried the previous day (because their sibling was reviewed that day), making it easier to guess the answer.

[dae]

When set to due order, if you search for is:due in the browse screen and order by the due column, you should find the cards that appear during review appear in the same order as the due dates shown there. If that’s not the case, please provide an example.

I’ll add an option to randomize reviews in the next update.

[vaibhav]

I’ll add an option to randomize reviews in the next update.

Thanks for that. It will most likely solve my second problem (buried cards from yesterday being shown first). It may also solve my first problem if it behaves in a manner similar to how the Random option in filtered deck “cards selected by” option does.

When set to due order,

By this, do you mean the “Due date, then random” option?

When set to due order, if you search for is:due in the browse screen and order by the due column, you should find the cards that appear during review appear in the same order as the due dates shown there.

I will assume this to be working as intended.

But, my problem is that the order is not truly random, despite the option name saying “Due date, then random”, which means that the order should be random if all the due cards have the same due date.

If all cards have the same due date, I have observed that the filtered deck “Random” option shows the cards in a truly random order (which I say because it doesn’t tend to club related cards together) while the “Due date, then random” option does tend to club the related cards together.

[dae]

I can’t seem to reproduce the behaviour you describe. The following in the debug console appears to show that the filtered and not filtered orderings are roughly equal in how far creation dates of cards are spaced out between consecutively-presented cards.

import time, statistics
for n in range(0, 10):
  for order in "random()", "fnvhash(id, mod)":
    last=0
    dists=[]
    rows = mw.col.db.all(f"select id, nid, ord from cards where due in (885, 886) order by {order}")
    for row in rows:
      id, nid, ord = row
      if last:
        dist=abs(last-id)/1000/60/60.0/24.0
        dists.append(dist)
      last=id  
    print(order, "median distance (days)", statistics.median(dists))

random() median distance (days) 61.94717258101852
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 59.955485613425935
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 43.9002203587963
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 57.90288971064815
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 59.20066923611111
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 55.44655893518518
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 50.9532762037037
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 54.00831753472222
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 51.959114236111105
fnvhash(id, mod) median distance (days) 45.00525592592593
random() median distance (days) 55.213709189814814
fnvhash(id, mod) median distance (days) 45.00525592592593

Here random tends to be slightly higher, but look forward two days and the opposite applies:

random() median distance (days) 121.9628469675926
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 108.10999886574075
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 116.9246264814815
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 86.03269311342592
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 100.85414918981483
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 103.04565670138889
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 110.96755107638889
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 110.97810988425925
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 102.0261784837963
fnvhash(id, mod) median distance (days) 120.89984241898149
random() median distance (days) 112.03415050925925
fnvhash(id, mod) median distance (days) 120.89984241898149

[vaibhav]

The following in the debug console appears to show that …

Sorry, I am not that technical to understand that output.

I can’t seem to reproduce the behaviour you describe.

In my experience, such behaviour is more likely to occur if I study for several days by clicking on the main deck.

When I study using a filtered deck for few days, then that behaviour is not seen for some days even after I switch back to clicking on the main deck. Some days later, related reviews again begin to appear together more often.

Because all this happens over a course of several days, I am unable to give specific examples. However, I do feel this to be happening and this does seem to be resolved by using filtered decks.

[vaibhav]

Is it possible that this issue results from the fact that I do my reviews on AnkiDroid?

I am using AnkiDroid v2.16alpha97 with the new backend and v3 scheduler enabled. However, I have come to realize that it doesn’t behave similar to Anki v2.1.61 in all aspects (though it does in most aspects).

A similar issue was reported here before the Review Sort Order setting was added to Anki: Maximally spaced filtered deck order

[dae]

If you suspect AnkiDroid, you could try avoiding it for a while and see if the problems go away.