Sort by R after postponing

[user1823]

Fixes #406

[user1823]

The postpone function uses fuzz. So, I have taken the average when estimating the retention to prevent excessive calculations.

[user1823]

In addition, I have replaced interval with stability in the second sort condition because stability is what actually decides how fast R falls.

[L-M-Sherlock]

What's the purpose of plus ivl * 0.075?

[Expertium]

I have the same question as well.

[user1823]

It is used to get an estimate of the elapsed days at the new due date.

The code for the new interval is this:

        delay = elapsed_days - ivl
        new_ivl = min(
            max(1, math.ceil(ivl * (1.05 + 0.05 * random.random())) + delay), max_ivl
        )

For the reason explained in #407 (comment), I assume this to be:

        delay = elapsed_days - ivl
        new_ivl = min(
            max(1, math.ceil(ivl * (1.05 + 0.025)) + delay), max_ivl
        )

If you put the value of delay, you get

        new_ivl = min(
            max(1, math.ceil(ivl * 0.075) + elapsed_days), max_ivl
        )

For the purpose of sorting, we can skip math.ceil and assume that the value is between 1 and max_ivl. So, we get new_ivl = ivl * 0.075 + elapsed_days

At due date, R = power_forgetting_curve(new_ivl, stability)

[Expertium]

Oh, interesting. That's a clever way to take fuzz into account.

[user1823]

assume that the value is between 1 and max_ivl

If you don't want to make this assumption, we can change it to something like

THEN min({mw.col.sched.today} - (due - ivl) + ivl * 0.075, max_ivl)

[Expertium]

Ok

[user1823]

The value not being > 1 is not actually a problem because it is elapsed_days + ivl * 0.075 and elapsed_days would always be at least equal to 1 (you can't postpone a learning card).

The main concern is that this value can sometimes become greater than the max_ivl but the add-on would not assign that value of interval. So, the retention estimate would become inaccurate.

[L-M-Sherlock]

LGTM