World: r3wp

A little more
>>echo %trace.txt
>>trace on
>>do %rebgui.r
>>echo none
Trace:  echo (word)
Trace:  none (word)
Trace:  if (word)
Trace:  port? (word)
Trace:  system/ports/echo (path)
Result: true (logic)
Trace:  [
    close system/ports/echo
    system/ports/echo: none
] (block)
Trace:  close (word)
Trace:  system/ports/echo (path)
REBOL Internal Error: Stack overflow


                                    Program terminated abnormally.

                                                                  This should never happen.

                                                                                           Contact www.REBOL.com with details.

Can't reproduce that under Win or Mac. Does the error occur if you 
'do another script and quit (i.e. is it dependent on loading %rebgui.r)? 
Failing that, try commenting out the third last line of %rebgui.r 
(system/view/face/feel: none) and see if that changes things.

Henrik - gradient being assymetric is what I like. Well, the only 
one trouble for me is small visual "defect", where buttons don't 
look too good on dark background, because they miss more concrete 
border. I wonder if it would look good, if buttons shared default 
widget border? Well, probably not :-)

But - visual look is area for customisation for anyone. However - 
I showed mine adapted default look to 3 ppl - 1 PC admin, 1 web designer, 
one reboller, and most agreed that the difference (using mild blue 
colors and grays, non-rounded shaping of tabs) is some 30% of feel 
to them :-)

hehe, just tried to make button and table tabbable, got following 
error - imo because of table :-)

** Script Error: empty? expected series argument of type: series 
port bitset
** Where: set-focus
** Near: view*/caret: case [

    all [caret in face 'caret face/caret] [at face/text face/caret]
    find behaviors/caret-on-foc...

segmentation fault at quit on Linux is very common with REBOL. I 
don't think it's rebgui related.

Ashley;  Yeah , this seems to be an internal stack problem in REBOL 
beta  Sorry for dropping this here.

Pekr, can you post your code so we can reproduce the error?

Cyphre's spell checker http://www.rebol.cz/~cyphre/spell.r

And Ladislav posts an improvement leading to 3x speed increase.

post your code

 ... I suspect it was as simple as adding a fiew widget names into 
 behaviors/tabbable via request-ui

posts an improvement leading to 3x speed increase
 ... where?

mailing list

hum, i'm not sure that all modifications made by ladislav are improvment

his 'train function is slowest than the primary

yes, it is, but it does more things

^^^

ladislav, i gan 1 second while intializing the dictionary with this 
train function:

train: func [
	words [block!]
	/local keys values pos i
][
	keys: to hash! unique words
	values: head insert/dup cp [] 0 2 * length? keys

 i: 0 loop length? keys [change values first keys values: skip values 
 2 keys: next keys]
	values: head values
	keys: head keys
	foreach word words [
			pos: 2 * index? find keys word 
			poke values pos 1 + pick values pos
	]
	sort/reverse/skip/compare values 2 2
	values
]

the first 'sort in your train function consume a lot of time

anyway, my function is not perfect ...either

The 'correct function is *very* slow, takes almost 4 seconds to do 
the following:

	>> correct "caustic"


Also, where is the bin file sourced from and are other languages 
supported?

yeah all languages are supported, it's the good thing with this algo

you juste have to replace the corpus by another bin file

i mean, all "latin" languages are supported

And a good collection of bin files would be located at?

hey, they are just plain text books

just take a book well writed

what is your language ?

take a classic

Queen's English ;)

use books.google.com to find a good one

I'm wondering whether, for RebGUI, this approach is actually "better" 
than the current soundex approach? If it means having to pick arbitrary 
literary works for each language then what's the point? I was hoping 
that someone had pre-generated some "standardized" dicts somewhere 
I could grab.

i dunno

soundex is faster

but more complicated

? It's less than a dozen lines of code!

oh really ?

right

http://trac.geekisp.com/rebgui/browser/rebgui-edit.r?format=raw

steeve: don't forget, that the 1s difference in TRAIN speed does 
not actually matter, since it is just a dictionary preparation. What 
matters is the speed of the CORRECT function

how fast is the correct function?

the version I just sent to the ML is now almost 4 times faster than 
Cyphre's

on his example it is about 9 words/s

Graham - Ashley is right - all I did wat I added 'table name to the 
behavior tabbing category ....

Soundex is a language-specific algorithm, unfortunately

So soundex is specific to English ?

Seems so: http://en.wikipedia.org/wiki/Soundex

Looking at the new algorithm again, a lot of effort goes into weighting 
word use (the values block) which in turn is only used in one place 
... to determine the "best" candidate replacement. Given that a spell-checker 
just needs to return a sorted candidate list and the remainder of 
the algorithm handles this quite well, I'd be quite happy with a 
cut-down version that excludes the whole dict/values logic. That 
way I could use the existing dictionary files I have sourced (sans 
soundex codes).

it is easy to cut down - just remove the WORDS and TRAIN functions 
and replace their use by reading the DICT data

So, to create a specialized dictionary, you can train it on eg. a 
medical text ?