Fix/wrong calc elapsed days

[ishiko732]

ref: https://forums.ankiweb.net/t/feature-request-estimated-total-knowledge-over-time/53036/58?u=l.m.sherlock

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 100.00%. Comparing base (af8f873) to head (848875c).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #144   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           15        15           
  Lines          689       693    +4     
  Branches        76        76           
=========================================
+ Hits           689       693    +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Luc-Mcgrady]

Before:

Now:

https://github.com/Luc-Mcgrady/Anki-Search-Stats-Extended/blob/line-graph/src/ts/MemorisedBar.ts
Provided my code works as intended, there are still cards which have their stabilities different between running it with this and Anki.
Could it be Anki's rollover?

[L-M-Sherlock]

Could it be Anki's rollover?

Could you provide a unit test to reproduce the inconsistency? Like this:

test('TS-FSRS-Simulator', () => {
    const f = fsrs({
        w: [1.1596,1.7974,13.1205,49.3729,7.2303,0.5081,1.5371,0.001,1.5052,0.1261,0.9735,1.8924,0.1486,0.2407,2.1937,0.1518,3.0699,0.4636,0.6048],
    })
    const rids = [1704468957000, 1704469645000, 1704599572000, 1705509507000]
    let card = createEmptyCard(new Date(rids[0]))
    const grades : Grade[] = [Rating.Good, Rating.Good, Rating.Good, Rating.Good]
    for (let i = 0; i < rids.length; i++) {
        const now = new Date(rids[i])
        const log = f.next(card, now, grades[i])
        card = log.card
        console.log(card.stability)
    }
  })

[ishiko732]

Provided my code works as intended, there are still cards which have their stabilities different between running it with this and Anki.
Could it be Anki's rollover?

Since I cannot directly run the Anki-Search-Stats-Extended code, I'm not sure if there's an issue with it; perhaps the last scheduling in historical cards was not using FSRS-5, but FSRS-4 or FSRS-4.5, which might explain the inconsistency in calculated stability?

[ishiko732]

some Anki card records have s values based on FSRS-4.5.

select *
from revlog
where cid= 1631872042742
order by id desc

#,id,cid,usn,ease,ivl,lastIvl,factor,time,type
1,1714744568062,1631872042742,612,3,74,31,861,60000,1
2,1711352380923,1631872042742,327,2,36,24,890,60000,1
3,1709171131116,1631872042742,150,3,20,10,815,7451,1
4,1708334378635,1631872042742,109,3,6,3,827,554,1
5,1708102403848,1631872042742,101,2,3,2,841,6217,1
6,1707902884484,1631872042742,93,2,2,2,763,46325,1
7,1707640884786,1631872042742,84,4,2,-60,682,3651,0
8,1707640515087,1631872042742,84,2,-330,0,682,27105,0

select *
from cards
where id= 1631872042742

id,nid,did,ord,mod,usn,type,queue,due,ivl,factor,reps,lapses,left,odue,odid,flags,data
1631872042742,1631872042742,1707630005524,0,1714744568,612,2,2,303,74,2050,8,0,0,0,0,0,"{""pos"":81,""s"":68.49,""d"":7.846,""dr"":0.9,""cd"":""{""v"":""v4.11.1"",""seed"":7621,""d"":7.1,""s"":70.85}""}"

---

ts-fsrs v4.5.2 using FSRS-5

test('TS-FSRS-Simulator[ts-fsrs v4.5.2 FSRS-5]', () => {
  const f = fsrs({
    w: [
      1.0139, 1.0139, 4.6959, 11.3189, 5.0903, 1.121, 0.8626, 0.0606, 1.628,
      0.1, 1.0006, 2.1484, 0.0825, 0.3375, 1.4004, 0.1114, 2.7801,
    ],
  })
  const rids = [
    1707640515087, 1707640884786, 1707902884484, 1708102403848, 1708334378635,
    1709171131116, 1711352380923, 1714744568062,
  ]
  const expected = [
    1.0139, 1.0139, 1.75389969, 2.25526707, 8.22141138, 26.63123583,
    31.49567267, 91.53053746,
  ]
  let card = createEmptyCard(new Date(rids[0]))
  const grades: Grade[] = [2, 4, 2, 2, 3, 3, 2, 3]
  for (let i = 0; i < rids.length; i++) {
    const now = new Date(rids[i])
    const log = f.next(card, now, grades[i])
    card = log.card
    expect(card.stability).toBeCloseTo(expected[i], 4)
  }
// card.stability => 91.53
})

---

ts-fsrs v3.5.7(patch elapsed_days) using FSRS-4.5

git checkout v3.5.7
pnpm install

After applying the modifications from this PR to the src/fsrs/scheduler.ts#SchedulingCard.constructor related code, proceed with testing.

    card.elapsed_days =
-      card.state === State.New ? 0 : now.diff(card.last_review as Date, "days");
    card.elapsed_days =
+      card.state === State.New ? 0 : dateDiffInDays(this.last_review, now);

test("TS-FSRS-Simulator[ts-fsrs v3.5.7(patch elapsed_days) FSRS-v4.5]", () => {
  const f = fsrs({
    w: [
      1.0139, 1.0139, 4.6959, 11.3189, 5.0903, 1.121, 0.8626, 0.0606, 1.628,
      0.1, 1.0006, 2.1484, 0.0825, 0.3375, 1.4004, 0.1114, 2.7801,
    ],
  });
  const rids = [
    1707640515087, 1707640884786, 1707902884484, 1708102403848, 1708334378635,
    1709171131116, 1711352380923, 1714744568062,
  ];
  const expected = [
    1.0139, 1.0139, 1.7317146, 2.18208164, 6.98174642, 21.80173316, 25.74674998,
    68.49768733,
  ];
  const col: number[] = [];
  let card = createEmptyCard(new Date(rids[0]));
  const grades: Grade[] = [2, 4, 2, 2, 3, 3, 2, 3];
  for (let i = 0; i < rids.length; i++) {
    const now = new Date(rids[i]);
    const log = f.repeat(card, now)[grades[i]];
    card = log.card;
    col.push(card.stability);
    expect(card.stability).toBeCloseTo(expected[i], 4);
  }
  console.log(card.stability); // card.stability => 68.49
});

[Luc-Mcgrady]

Wow thanks for all the effort!

Reverting to v3.5.7 didn't work for me.

v3.5.7+fix:

This pr:

const f = fsrs({
    w: [2.3087, 4.1277, 9.8679, 17.6430, 7.2056, 0.3775, 1.7137, 0.0011, 1.4670, 0.1637, 0.9486, 1.9157, 0.1287, 0.3025, 2.3155, 0.0001, 2.8344, 0.3571, 0.6384]
  })
const rids = [1665423581318,1665423653089,1665424320968,1665433930433,1665480844915,1665749731632,1666458872005,1667929850092,1671571024108,1680131629095,1689548964502]
const grades: Grade[] = [1,3,3,3,3,3,3,3,3,2,3]

I'd say that's close enough for the graph.

---

Here's an example problem card, though I expect this is out of the scope of this pull request.

id	cid	usn	ease	ivl	lastIvl	factor	time	type
1735993129140	1662942994437	-1	0	882	25	589	0	5
1713000017248	1662942994437	18391	3	624	119	589	6207	1
1704910583686	1662942994437	16287	0	143	143	590	0	4
1702718934003	1662942994437	15532	3	102	47	565	2668	1
1699260169343	1662942994437	14468	0	20	18	550	0	4
1698698192380	1662942994437	14275	3	18	-7200	2500	4235	0
1698688916440	1662942994437	14271	3	-7200	-600	0	1326	0
1698688837021	1662942994437	14271	3	-600	-60	0	1363	0
1698688771401	1662942994437	14271	1	-60	-7200	0	6472	0
1698678126399	1662942994437	14261	3	-7200	-600	0	1965	0
1698678054940	1662942994437	14260	3	-600	0	0	10617	0
first_of_last_revlog_entries	^	so I truncate it here
1697182240703	1662942994437	13788	0	0	139	0	0	4
1688803646964	1662942994437	9846	3	100	70	2450	3962	1
1682408696385	1662942994437	7589	3	129	53	2450	9448	1
1677870162608	1662942994437	5711	3	53	27	2450	21773	1
1673944811781	1662942994437	4498	3	102	42	2450	8779	1
1670319241213	1662942994437	3546	4	42	14	2450	4508	1
1669104703673	1662942994437	3216	3	14	6	2300	18944	1
1668610746240	1662942994437	3031	3	6	-86400	2300	4327	2
1668507588958	1662942994437	2985	3	-86400	-600	2300	3370	2
1668506742137	1662942994437	2984	3	-600	-60	2300	5112	2
1668506506324	1662942994437	2983	1	-60	-86400	2300	30000	2
1668429286472	1662942994437	2946	3	-86400	-600	2300	9298	2
1668427317488	1662942994437	2943	3	-600	-60	2300	3325	2
1668427235731	1662942994437	2941	1	-60	27	2300	12559	1
1666095772446	1662942994437	1738	3	27	11	2500	4827	1
1665140780850	1662942994437	1106	3	11	5	2500	7384	1
1664707697911	1662942994437	705	3	5	3	2500	20502	1
1664439011033	1662942994437	563	3	3	1	2500	4657	1
1664392046978	1662942994437	548	3	1	-7200	2500	12497	0
1664382514443	1662942994437	542	3	-7200	-600	0	5423	0
1664382303688	1662942994437	540	3	-600	-60	0	4676	0
1664382285359	1662942994437	540	1	-60	-7200	0	8144	0
1664375482797	1662942994437	533	3	-7200	-600	0	17122	0
1664371293852	1662942994437	528	3	-600	-60	0	6710	0
1664371215663	1662942994437	528	1	-60	-60	0	9852	0

Card:

id	nid	did	ord	mod	usn	type	queue	due	ivl	factor	reps	lapses	left	odue	odid	flags	data
1662942994437	1662942994329	1663460765931	1	1736029349	-1	2	2	684	103	2500	8	0	0	0	0	0	{"pos":7916,"s":71.77,"d":8.23,"dr":0.87,"cd":"{"v": "reschedule"}"}

v3.5.7+fix:

This pr:

Heres the card quickly pasted into FSRS-rs

test('SSE', () => {
  const f = fsrs({
    w: [0.4911, 4.5674, 24.8836, 77.0450, 7.5474, 0.1873, 1.7732, 0.0010, 1.1112, 0.1520, 0.5728, 1.8747, 0.1733, 0.2449, 2.2905, 0.0000, 2.9898, 0.0883, 0.9033]
  })

  const rids = [1698678054940,1698678126399,1698688771401,1698688837021,1698688916440,1698698192380,1699260169343,1702718934003,1704910583686,1713000017248]
  const ratings: Rating[] = [3,3,1,3,3,3,0,3,0,3]

  let card = createEmptyCard(new Date(rids[0]))

  for (let i = 0; i < rids.length; i++) {
    const rating = ratings[i]
    if (rating == 0) {
      continue
    }

    const now = new Date(rids[i])
    const log = f.repeat(card, now)[rating]
    card = log.card
  }


 // For pasting into ts-fsrs
  let last = new Date(rids[0]);
  let delta_t: any[] = []
  for (let i = 0; i < rids.length; i++) {
    const rating = ratings[i]
    if (rating == 0) {
      continue
    }

    const current = new Date(rids[i]);
    delta_t.push({delta_t: dateDiffInDays(last, current), rating})
    last = new Date(rids[i]);
  }
  console.log(delta_t)
  
  expect(card.stability).toBeCloseTo(71.77)
})

[L-M-Sherlock]

So, is it OK to merge this PR?

[L-M-Sherlock]

LGTM

[ishiko732]

Here's an example problem card, though I expect this is out of the scope of this pull request.

Let's discuss how to handle this special case in issue #145.