Ordering Request: Reverse Relative Overdueness

Anki Suggestions
207

It would have to be two conditions, that and no new cards left either.

208

@L.M.Sherlock So I have a few things that I see that (I think) need to be changed on the sim and it’s no longer relevant to this discussion. Should I create a new thread on the anki forum, or is there another place you’d prefer I do that, like on github itself? I don’t have a github, so I’m new to that.

209

You can open an issue on Github: GitHub - open-spaced-repetition/review-sort-order-comparison

210

Oh yeah, check out how slowly due_date_asc is going through that backlog after the new cards ran out. The whole sim is going to take a while, but I can already see this will make a huge difference.

Screen Shot 2024-09-30 at 3.04.40 PM
Screen Shot 2024-09-30 at 3.04.40 PM448×571 67.6 KB

211

Ok, so here’s two to compare. This is gonna take forever and I’m going to leave and let it sit. I’ll post all the results later.

Screen Shot 2024-09-30 at 3.30.41 PM
Screen Shot 2024-09-30 at 3.30.41 PM430×591 15.4 KB

Screen Shot 2024-09-30 at 3.35.58 PM
Screen Shot 2024-09-30 at 3.35.58 PM446×573 15.4 KB

212

I lied. I stopped the sim and just ran PRL because I wanted to see it now:

Screen Shot 2024-09-30 at 3.46.54 PM
Screen Shot 2024-09-30 at 3.46.54 PM428×520 14.5 KB

Ok, now I’m leaving and letting it run them all.

213

What sorting method are you testing here

214

Last time.

Screen Shot 2024-09-30 at 4.04.28 PM
Screen Shot 2024-09-30 at 4.04.28 PM431×544 14.4 KB

Screen Shot 2024-09-30 at 4.04.51 PM
Screen Shot 2024-09-30 at 4.04.51 PM437×575 15 KB

Screen Shot 2024-09-30 at 4.05.08 PM
Screen Shot 2024-09-30 at 4.05.17 PM
Screen Shot 2024-09-30 at 4.05.40 PM

215

It says under the last day for each one, when it gives the final stats.

216

It seems to confirm that Retrievability Descending is best sorting method overall.

217

No, I think it’s confirming PRL is. Retrievability Descending takes the longest of those three.

218

Did you include the changes you wanted to do earlier (making the simulation run for longer) :question:

219

Yeah, this is running the sim until there’s no backlog. So, on the last day your 80 reviews covers every review you have left to study that day.

220

Ah great! But it is still hard to determine a clear winner here. What does PRL stand for btw and is it hard to add to Anki as a sorting method :question:

Dae has already added the inclusion of Reverse Relative Overdueness to the next as an issue on Github.

221

Potential Retrievability Loss (PRL) = R(Today) - R(Tomorrow)

222

I think total time represents a good winner. Total minutes of studying time it took to completely learn all your cards and get through your backlog.

223

Alright, I did it the dumb way - let’s add 1000 to learn_span and just look the graph, without finding the exact day when average true retention matched desired retention. The problem is that the graph has 16 curves, which makes it hard to tell what’s going on

Figure_2
Figure_21616×775 172 KB

Figure_4
Figure_41551×705 166 KB

224

Kinda weird that difficulty_asc plummets in True Retention per Day as soon as you run out of new cards.

225

I was worried about Difficulty Ascending because Difficulty is not dynamic (changing with days passed) like Retrievability (on which PRL and Retrievability Descending obviously depend).

Maybe that has an influence.

226

It appears that in your simulations Retrievability Descending also remains to be the winner.

@rich70521 the weird plummet of Difficulty Ascending is also consistent in Expertiums simmulations.