Interesting. I did not think about this approach before, maybe that's the correct way to go.
I would like to use some control loop protocol that found to be robust on Internet, MDAI (Multiplicative Decrease, Additive Increase).
I think I need to separate the notion of order and frequency. Frequency may be used to generate the initial order, but order can be changed at run-time. They are stored separately. Like what you mentioned, order can get promoted or punished once the predicition is hit/missed. Pubishment should be severe than promotion. (say, once selected, a lexicon will be promoted to the first order, otherwise, the order degenerate everytime)
However, this increase the space complexity of this database a lot. Is it worth to do this? I think yes.
// Shawn
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
PJk (http://members.xoom.com/ezinput) @M GPL QUwC
~ABBS.cis.nctu.edu.tw Chinfost ذϪ [ (words)/C (word lists)] ؿ̡A]@ǨLwTC
٦ApG@ǴzJkΤAWǰ^m@ǹq~PӻݤʿrɡAӤ]OӤwӷC
> PJk (http://members.xoom.com/ezinput) @M GPL QUwC
libtabe ثeQ|UwCƹW쥻ӦbQUkAL���\h~観DAzLHuβέpkRFCAiH@Uoӵw|ʦpAƦܬݤ@UPJkoQUr~C
> ~ABBS.cis.nctu.edu.tw Chinfost ذϪ [ (words)/C (word lists)] ؿ̡A]@ǨLwTC
ڷQӦC ctsai4 (ӯE) COP@AڴNOqLRLwARC
> ٦ApG@ǴzJkΤAWǰ^m@ǹq~PӻݤʿrɡAӤ]OӤwӷC
oOثeݨӳ̦nkAL٨SۤWC
Thanks.
// Shawn
libtabe MPJkwtC
http://william.cswiz.org/techreport/diff-tabe-ez.txt.gz ] 648KB^
ɮײͨBJG
1. X libtabe 0.1 src/tsi-src/tsi.src ɪĤ@C
ƨӷGhttp://libtabe.sourceforge.net/
2. N ezbig.cin ɪOh]H % }Y^, XĤG,
ѤpjƧ, hƦ, AoŸ]p j ^B
^r]p ݡBVisual Basic^AXFC
ƨӷGhttp://members.xoom.com/ezinput/
3. H Unix diff ROɪtC
䤤, Hu<v}YUO libtabe BPJkLF
Hu>v}YhOPJkBlibtabe LC
BέpG
1. libtabe 0.1 w, @p 138,650 ]tr^F
PJkw, @p 103,986 ]tr^C
2. libtabe BPJkL, @p 63,244 F
PJkBlibtabe L, @p 28,581 C
3. PJktYǵ, ȤAXiӪ libtabe ,
pGu@vBu@~vC
ǵ], pGu@ѧ@v]ӬOu@@v^C
Thanks.
ݰ_ӧA]٦NӵwX_? 8)
libtabe N ctsai4 wP xcin iome X_ӮɴNoӰD, ɬOzLXӺͥΤu覡, N���@ǿ~DX. (ڰOoݤF§j]ݮth@bӤw, ǤHݨY, ǤHΫeݤ@IRv) ... ӴN, εWhR@Ъ~, p@@, ]W|Ӱ, Dja. OooӤӦn, ]aw. 쥻X_ӪwOoNQU, ڧҦWӧC, AǤuR, ooӤjp. (WL{b tsi-src ̭)
DH@Nf\w, _h merge |~U. ӥB, bSWpU, ]|ӦγB.
// Shawn
ڤ@{, έpk, ٬OAXUγ~;
¦֤ߵw, ٬OӥΤHuC
ڤߥؤ{̲zQw, NOmШ|ynG
http://www.edu.tw/mandr/clc/dict,
iOqתw, L,
Y]ֵuX{bMpC
M http://www.edu.tw/mandr/clc/dict/htm/s/sc.htm һ,
ubgLӽг\iA}UɱqƫDӷ~ت
dzNsBШ|sγ~ϥΡCv
pGSkov, ܦۤvg robot,
Ρu^xvd, ]W@Uhj... II... :q
oصwFk٬OIvC 8)
PJk}ӵwOiHݤ@UCLıoNثep
AwAjȤ]ɤFh֥TvC`εjaAjwh
XӪhbOCWΨuΦyAUqjCثe@kDn~
VOLkS@ӾhBzsAҦ{XӪhrr곣|Q
���ѳrզrCpGBzsANƤFDzߥ\
FC
> LıoNثep AwAjȤ]ɤFh֥TvC
> `εjaAjwhXӪhbOCWΨuΦyAUqjC
S, oǵwGO@٦~, MLˤj.
>ثe@kDn~VOLkS@ӾhBzsAҦ{XӪhr
>r곣|Q���ѳrզrCpGBzsANƤFDz
>FC
ƹWڰLo譱VO, ƱiHbSHuѻPpUͷs, L٬OѤF.
ɪkOЧQΤjyƮw, ϥμɤOkNQӦrHUҦƦCզXέpX.
o˧ڭ̱o쪺O@ӦrWv. qrWv, ڳ]pF@ herustics, Ҧp,
pG@ӥ|rO@Ӥr prefix suffix, W۶Zb 10% M϶, h
iHNrR (o, ܤ֬O@ӨҤl). heuristics T, Oo
ƦCզXbӹLɤO, Ӧh. LoӤkӥiHΦb xcin ϥΪ feedback
module , @譱ϥΪ̪JeDz, @譱, ϥΪ̤]iH㪺.
ګOͷsAӬOuW{s]Φ^CҦp|
DzߪtΡAhbu|ϥΪ̿J cai4 zhi4 hao4 নu
v]ܪWvjӡAWvjE^AMCCpG
uWDzߡAϥΪ̤ʪקOUӡA[Jw]iHOw
]iHOʺAϥΪ̵w^AĤGAJP˪JAN|A
FC
FѡAo]O{q@Q xcin community push @n\CO@ ideaAL٨S implementC
idea ²A{b xcin wgѤ@ prompt @ input sequence _ˤlAҦpuClibtabe iHǥѦ^oǸơAs@@ user preference databaseCodzO implementationA²C
OAoUlӵwAƦܧhA_ɭnӵwW[_ӬݩO? ٬O user preferenceArOOAέ즳wh_Ao˷ user preference Ӥ֮ɡAXvi|ܰC
ǥѦ^o user preferenceAiHΨ adapt µweC
wbjAǫhFC̪b yyj-winAMOβέp
disambiguation ӤOw說C{uRAѡv
Ӧh]åΫܤjyƥhVm^ADnOaϥΪ̪JӾDz
CҥH@}l~vA@qɶN۷���FC
說 libtabe ӻAiNOW[BzʺAw
Fo{sBxssBηsCܩsvDA
ڷQDz߱oӡC{Ĥ@gѨϥΪ̦^XǨYӷsɡA
@ӤʪvCH{|_ϥΪ̪sAϥΪ̥h勵AN
vC{_XsAϥΪ̤nANէCvC
Interesting. I did not think about this approach before, maybe that's the correct way to go.
I would like to use some control loop protocol that found to be robust on Internet, MDAI (Multiplicative Decrease, Additive Increase).
I think I need to separate the notion of order and frequency. Frequency may be used to generate the initial order, but order can be changed at run-time. They are stored separately. Like what you mentioned, order can get promoted or punished once the predicition is hit/missed. Pubishment should be severe than promotion. (say, once selected, a lexicon will be promoted to the first order, otherwise, the order degenerate everytime)
However, this increase the space complexity of this database a lot. Is it worth to do this? I think yes.
// Shawn
What I have suggested was a simple mechanism of reinforcement learning. Shouldn't be too difficult to implement.