Re: some news / ideas

On Sunday 05 October 2003 18:47, David Calinski wrote:

> Yes, exactly - that's why this idea is not so great. :)
> But on first reviews (especially first and second) this data ("guessed
> difficulty") doesn't work so well.
> On first presentation elements of "high" difficulty could be scheduled
> closer. 

I avoid this problem altogether by using the 'drill badly known' mode until 
the items have grade 4. After that, they can go into the normal scheduled 
revision process, and the proposed schedules then don't seem too bad.

I could be wrong, but I see the main benefits of scheduled revision in keeping 
the memories 'alive' rather than acquiring them. For initial acquisition I 
much prefer 'drill badly known'.

> Maybe indeed, instead of implementing it and bloating code and everything,
> there should be a good manual and user should know how and what grades
> should choose.

Yeah, that's right. I also think it's important to stress that users should be 
consistent in the grades they choose. I guess the NN will not be very good if 
the user gradually changes his/her meaning of e.g. grade 3 over time.

Now that we're on that subject, I'm not sure how useful it would be to train 
the NN on data from lots of different users, because my interpretation of the 
grades could be different from yours. This is fine, as long we're each 
consistent in how we grade ourselves.

So the NN just needs time to adapt to our own personal learning profile and 
grade use.

Cheers,

Peter

------------------------------------------------
Peter Bienstman
Ghent University, Dep. of Information Technology
Sint-Pietersnieuwstraat 41, B-9000 Gent, Belgium
tel: +32 9 264 34 45, fax: +32 9 264 35 93
WWW: http://photonics.intec.ugent.be
email: [email protected]
------------------------------------------------



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf