Question concerning the algorithm

[Henrik in Oslo]

I have for 3-4 weeks not included a couple of tags in my daily learning. I checked the effect of different gradings on some cards and was very surprised that "Easiness" remains practically unchanged whatever grade I give to these cards. (I need to understand this in order to manipulate txt import from Supermemo 7 which I can do in bulks based on "Easiness". In Mnemosyne they will of course start off as Easiness 2.5 and days-to-first repetition will be determined by grade given at first learning). A minor comment: The terminology is not altogether the same between "Show statistics" (Ctrl T) and "Browse Cards"  

Card #1 Before/ After

Grade: 5/5

Easiness: 2.56/ 2.66

Repetitions (Review reps): 6/ 6

Lapses: 0/ 0

Scheduled interval (?): 6/ ?

Last repetition: 25 days ago/ today

Next repetition: 18 days ago/ in 2 months

Average thinking time (secs): 0/ ?

Total thinking time (secs): 0 / ?

Created: 06/10/12

(Learning reps)

Card #2 Before/ After

Grade: 5/ 0

Easiness: 2.46/ 2.46 

Repetitions (Review reps): 6/ 6

Lapses: 0/ 1

Scheduled interval (?): 6/ ?

Last repetition: 25 days ago/ today

Next repetition: 18 days ago/ tomorrow

Average thinking time (secs): 0

Total thinking time (secs): 0

Created: 06/10/12

(Learning reps)

Thanks in advance

[Peter Bienstman]

Quoting "Henrik in Oslo" <001h...@gmail.com>:

> I have for 3-4 weeks not included a couple of tags in my daily learning. I
> checked the effect of different gradings on some cards and was very
> surprised that "Easiness" remains practically unchanged whatever grade I
> give to these cards.

Well, with grade 5 you do see an increase of easiness of 0.10 :-) For
failing grades, the easiness is indeed not changed.

All of this is exactly as prescribed by the original SM2 algorithm.

> A minor
> comment: The terminology is not altogether the same between "Show
> statistics" (Ctrl T) and "Browse Cards"

Could you please elaborate?

Thanks!

Peter

[Henrik in Oslo]

Hi again

A Easiness/ how the E-factor of a given item is modified

"Well, with grade 5 you do see an increase of easiness of 0.10 :-) For  failing grades, the easiness is indeed not changed. All of this is exactly as prescribed by the original SM2 algorithm" 

I fail to see that for card 2 this is according to SM-2 algorithm. 

I find a description of the SM-2 algorithm here 

http://www.supermemo.com/english/ol/sm2.htm

It says: 

«EF':=EF+{0.1-(5-q)*[0.08+(5-q)*0.02]}

where:

EF' - new value of the E-Factor,

EF - old value of the E-Factor,

q - quality of the response in the 0-5 grade scale.

If EF is less than 1.3 then let EF be 1.3.»

In my "card 2" example

EF = 2.46

q = 0

I)      EF':=EF+{0.1-(5-q)*[0.08+(5-q)*0.02]}

II)     EF':=2.46+{0.1-(5-0)*[0.08+(5-0)*0.02]}

III)    EF':=2.46+{0.1-(5)*[0.08+(5)*0.02]}

IV)   EF':=2.46+{0.1-(5)*[0.08+0.1]}

V)    EF':=2.46+{0.1-(5)*[0.18]}

VI)   EF':=2.46+{0.1-0.9}

VII)  EF':=2.46+{-0.8}

VIII) EF' = 1.66

(Excuse all the details and step, but I haven't done maths for some years so I need all the steps to understand where we differ in interpretation of algorithm :-) )

For card 2 with old E-factor = 2.46 I expect new value of E-factor to become 1.66. 

1) Where do I go wrong?  

2) Under what circumstance does the E-factor decrease?

. 

B Terminology:

1) Is "Repetitions" (in Stat) same ac  "Review reps" (in Browse Cards)?

2) In Card Statistics I find "Scheduled interval". What is the equivalent in Browse Cards? 

Best regards

Henrik

[Peter Bienstman]

On Sunday, August 05, 2012 10:55:46 AM Henrik in Oslo wrote:

> It says:
> «EF':=EF+{0.1-(5-q)*[0.08+(5-q)*0.02]}

That formula is only valid for pass grades, not fail grades. From the SM
website:

6. If the quality response was lower than 3 then start repetitions for the
item from the beginning without changing the E-Factor (i.e. use intervals
I(1), I(2) etc. as if the item was memorized anew).

But I agree that it's confusingly written.

> *2) Under what circumstance does the E-factor decrease?*

For grades 2 and 3.

> *B Terminology:*

> 1) Is "Repetitions" (in Stat) same ac "Review reps" (in Browse Cards)?

Because of lack of space, I abbreviated repetitions as reps. In Stats, reps
currently is learning reps + review reps, but I agree it's inconsistent with
the browser, so I've split this up by listing learning reps and review reps
separately in the statistics page too.

> 2) In Card Statistics I find "Scheduled interval". What is the equivalent
> in Browse Cards?

Scheduled interval is next rep - last rep. This indeed probably does not add
too much, so I've removed this from the statistics page.

Thanks for spotting this!

Peter

>
> Best regards
> Henrik
>
> On Sunday, 5 August 2012 17:02:23 UTC+2, Peter Bienstman wrote:
> > Quoting "Henrik in Oslo" <001h...@gmail.com>:
> > > I have for 3-4 weeks not included a couple of tags in my daily learning.
> >
> > I
> >
> > > checked the effect of different gradings on some cards and was very
> > > surprised that "Easiness" remains practically unchanged whatever grade I
> > > give to these cards.
> >
> > Well, with grade 5 you do see an increase of easiness of 0.10 :-) For
> > failing grades, the easiness is indeed not changed.
> >
> > All of this is exactly as prescribed by the original SM2 algorithm.
> >
> > > A minor
> > >
> > > comment: The terminology is not altogether the same between "Show
> > > statistics" (Ctrl T) and "Browse Cards"
> >
> > Could you please elaborate?
> >
> > Thanks!
> >
> > Peter

--
Peter Bienstman
Ghent University, Dept. of Information Technology
Sint-Pietersnieuwstraat 41, B-9000 Gent, Belgium
tel: +32 9 264 34 46, fax: +32 9 264 35 93
WWW: http://photonics.intec.UGent.be
email: Peter.B...@UGent.be

[Henrik in Oslo]

Supermemo ver 7 1995 certainly decreases E-factor also for fail grade (Score 0, 1, 2 and 3) 

To me this seems senisble. 

I made a test-base in Supermemo 7 and manipulated the PC clock.The below table shows the development of E-factor in SM7 (1995). 

(Repetition after 1 month (M1), 2 months etc.) 

Self Rating (q) = 0: M1: 2.075 M2: 1.650 M3: 1.300 M4: 1.300 M5: 1.300 M6: 1.300

Self Rating (q) = 1: M1: 2.075 M2: 1.650 M3: 1.300 M4: 1.300 M5: 1.300 M6: 1.300

Self Rating (q) = 2: M1: 2.180 M2: 1.860 M3: 1.540 M4: 1.300 M5: 1.300 M6: 1.300

Self Rating (q) = 3: M1: 2.360 M2: 2.360 M3: 2.360 M4: 2.220 M5: 2.220 M6: 2.220

Self Rating (q) = 4: M1: 2.500 M2: 2.500 M3: 2.500 M4: 2.500 M5: 2.500 M6: 2.500

Self Rating (q) = 5: M1: 2.600 M2: 2.600 M3: 2.600 M4: 2.600 M5: 2.600 M6: 2.700

As you can see, the lower the Self Rating the lower the E-factor, also for fail grade (Score 0, 1, 2 and 3 in SM)  

If during a repetition session in SM items are scored 3 or lower, these cards are saved for what is called "final drill". During "final drill" the scoring is "binary": Pass = 4 (or higher). Fail = 3 (or lower). I quite like the "final drill"-concept and would like to see it also in Mnemosyne: the possibility of really drilling the items that you by that repetition haven't yet learned well. 

[Peter Bienstman]

On Monday, August 06, 2012 03:01:13 PM Henrik in Oslo wrote:
> Supermemo ver 7 1995 certainly decreases E-factor also for fail grade
> (Score 0, 1, 2 and 3)

I guess SM-7 is different than SM-2, then :-)

I wouldn't dare to say which one is better without a detailed statistical
analysis, though.

> I quite like the "final drill"-concept and would like to see it also
> in Mnemosyne:

But it is in Mnemosyne, it's when you relearn the cards which you graded 0 or
1 during your scheduled reps :-)

The 'final drill' is seamlessly woven in your learning process, without
requiring users to know about yet another concept ('final drill') or
complicating the UI. This is very much the Mnemosyne philosophy.

Hope this explains things somewhat.

Thanks for the feedback!

Peter