1. AI Project
- Posted by Michael Sabal <mjs at OSA.ATT.NE.JP>
Feb 21, 1998
-
Last edited Feb 22, 1998
For everyone who wanted the font package, vector graphics package, or AI routine
s, you can now find them on my home-page.
Surf to http://home.att.ne.jp/gold/mjs/Grabbag/Grabbag.html
If you have any comments, suggestions, or queries about them, feel free to let m
e know personally or via the mailing list.
2. Re: AI Project
>This is because Euphoria applies the math operation to each element of =
>the sequence.
>This works because the default color (black) remains 0 with the =
multiply. >If you wanted a different background color, you could be =
sneaky:
>
> -- convert the 1's to the foreground color (fc)
> -- convert the 0s to the background color (bc)
> s =3D (s * (fc-bc)) + bc
>
>2. You can use pixel to display a sequence of pixels, and replace the =
>inner loop:
>
> for i =3D 1 to length( s ) do
> pixel( s[ctr], {x,y+i-1})
> end for
>
>3. You can replace the whole loop with display_image:
>
> display_image( {x,y}, s )
>
>so the final routine could be:
>
> global procedure tsi_text_a(atom x, atom y, atom c, object s)
>
> =20
> if atom(s) then
> s =3D {s}
> end if
> display_image( {x,y}, s*c )
>
> end procedure
>
>-- David Cuny
David,
That's a good idea, and if I didn't care about saving the background =
information, I'd use it. I might even add the above procedure to the =
font to allow users the option. However, my original idea was to make a =
font that didn't overwrite the background (e.g. image) with blank =
colors. The trouble I had with a lot of the fonts available for =
Euphoria was that they in white space (the middle of "O", for example) =
they filled it in with a background color. In some of my applications, =
that wouldn't work, because I wanted to see what was behind the letter, =
like if I'm captioning a photo on the photo. In some cases it's hard to =
see the letters, which might make a blackout option a good idea anyway. =
Thanks.
Mike
3. Re: AI Project
- Posted by dm31 at uow.edu.au
Nov 06, 2002
Well, first thing first. We should really get a GOOD N-Net Lib going
first.
Now, this is a thought about programming AI that I have had for
awhile. Everyone that tries programming AI uses a 'static' N-Net, by
that I mean, all the wiring is pre-programming and doesn't change,
connections change, but can not be made. Just look at how brains grow
thou. During the time of most develop, the brain GROWS connections. If
it learns how to catch a ball, it forms new links, changed the N-Net
topography. Sure, adults learn, which are normally just changing
connection values, but the really learning happens during childhood
and at a much-increased pass too. I believe to create successful AI we
would need to do something similar.
So how bout, the EU community be the first to create a truly
'evolving' N-Net.
Well, just my opinion anyways.
Cheers,
Dan McG
4. Re: AI Project
On 7 Nov 2002, at 10:22, dm31 at uow.edu.au wrote:
>
> Well, first thing first. We should really get a GOOD N-Net Lib going
> first.
>
> Now, this is a thought about programming AI that I have had for
> awhile. Everyone that tries programming AI uses a 'static' N-Net, by
> that I mean, all the wiring is pre-programming and doesn't change,
> connections change, but can not be made.
Until the shared memory libs was written, you couldn't change any Eu
program either. You couldn't execute sequences as if they were Eu code,
except in David Cuny's interpreter. We still can't launch threads, but parallel
processing without real threads may very well be an enormous blessing in
disguise. The other problem is longevity, winOSs haveto get rebooted
occasionally, and getting the program to remember what it knows and how it
knows, and what to do with what it knows, and the last state of the "brain",
will be a pain, because (for example) we cannot look up the list of vars
(names and contents) like Mirc or Lua can do.
Kat
5. Re: AI Project
Hi Kat,
this why disk files and databases were invented. I do this sort of thing every
day. Why do you think
it is hard to do?
7/11/2002 11:02:29 AM, Kat <kat at kogeijin.com> wrote:
>
> The other problem is longevity, winOSs haveto get rebooted
>occasionally, and getting the program to remember what it knows and how it
>knows, and what to do with what it knows, and the last state of the "brain",
>will be a pain, because (for example) we cannot look up the list of vars
>(names and contents) like Mirc or Lua can do.
>
---------
Cheers,
Derek Parnell
6. Re: AI Project
On 7 Nov 2002, at 11:09, Derek Parnell wrote:
>
> Hi Kat,
> this why disk files and databases were invented. I do this sort of thing every
> day. Why do you think it is hard to do?
Without using associated lists, can win32lib save any of it's vars *without
explicitly coding the var names in a save routine* to the disk files? Can it
trace the program flow, save that flow to disk, and after a reboot, initiate
back
to the state in the program it was when it was shut down, so all the returns
in the functions execute properly?
Kat
> 7/11/2002 11:02:29 AM, Kat <kat at kogeijin.com> wrote:
>
> >
> > The other problem is longevity, winOSs haveto get rebooted
> >occasionally, and getting the program to remember what it knows and how it
> >knows, and what to do with what it knows, and the last state of the "brain",
> >will be a pain, because (for example) we cannot look up the list of vars
> >(names
> >and contents) like Mirc or Lua can do.
> >
>
> ---------
> Cheers,
> Derek Parnell
>
>
>
>
7. Re: AI Project
7/11/2002 11:29:16 AM, Kat <kat at kogeijin.com> wrote:
>
>On 7 Nov 2002, at 11:09, Derek Parnell wrote:
>
>>
>> Hi Kat,
>> this why disk files and databases were invented. I do this sort of thing
>> every
>> day. Why do you think it is hard to do?
>
>Without using associated lists, can win32lib save any of it's vars *without
>explicitly coding the var names in a save routine* to the disk files? Can it
>trace the program flow, save that flow to disk, and after a reboot, initiate
>back
>to the state in the program it was when it was shut down, so all the returns
>in the functions execute properly?
>
Of course not, Kat. It was never designed to do any snapshot checkpointing. And
without built-in
help from Euphoria, this would be an expensive operation. One could design a
Euphoria application
that did this though, but the overheads would probably not make it worth while.
But why would we need this type of functionality? If it was really important to
pick up EXACTLY
where the user was at the last crash, it would be better served with a
fault-tolerant system. The
context we are talking about is a AI-like application used as a learning
exercise for the coders. We
are not trying to create a positronic life-support system for intersteller
travellers. Emulating a
tiny worm is going to stretch our capabilities as it is!
>
>> 7/11/2002 11:02:29 AM, Kat <kat at kogeijin.com> wrote:
>>
>> >
>> > The other problem is longevity, winOSs haveto get rebooted
>> >occasionally, and getting the program to remember what it knows and how it
>> >knows, and what to do with what it knows, and the last state of the "brain",
>> >will be a pain, because (for example) we cannot look up the list of vars
>> >(names
>> >and contents) like Mirc or Lua can do.
>> >
>>
>> ---------
>> Cheers,
>> Derek Parnell
>>
>>
>
>
>
---------
Cheers,
Derek Parnell
ICQ# 7647806
8. Re: AI Project
On 0, Kat <kat at kogeijin.com> wrote:
>
> On 7 Nov 2002, at 10:22, dm31 at uow.edu.au wrote:
>
> >
> > Well, first thing first. We should really get a GOOD N-Net Lib going
> > first.
> >
> > Now, this is a thought about programming AI that I have had for
> > awhile. Everyone that tries programming AI uses a 'static' N-Net, by
> > that I mean, all the wiring is pre-programming and doesn't change,
> > connections change, but can not be made.
>
> Until the shared memory libs was written, you couldn't change any Eu
> program either. You couldn't execute sequences as if they were Eu code,
> except in David Cuny's interpreter. We still can't launch threads, but
> parallel
> processing without real threads may very well be an enormous blessing in
> disguise. The other problem is longevity, winOSs haveto get rebooted
> occasionally, and getting the program to remember what it knows and how it
> knows, and what to do with what it knows, and the last state of the "brain",
> will be a pain, because (for example) we cannot look up the list of vars
> (names and contents) like Mirc or Lua can do.
>
> Kat
>
Wow. You sound almost like a Dredge fan.
Seriously, the "simulated" neural net is mostly data structures.
The hard-coded stuff will be the initial state of the data strucutres
and the Eu code itself which actually maniplulates the data structures.
I once borrowed a neural net lib and genetic algorithm lib from the
archives
to do something like this, but I failed. (And i dont have that code
anymore
either, in case youre wondering.)
jbrown
Linux User:190064
Linux Machine:84163
--
9. Re: AI Project
On 7 Nov 2002, at 6:59, jbrown105 at speedymail.org wrote:
>
> On 0, Kat <kat at kogeijin.com> wrote:
> >
> > On 7 Nov 2002, at 10:22, dm31 at uow.edu.au wrote:
> >
> > >
> > > Well, first thing first. We should really get a GOOD N-Net Lib going
> > > first.
> > >
> > > Now, this is a thought about programming AI that I have had for
> > > awhile. Everyone that tries programming AI uses a 'static' N-Net, by
> > > that I mean, all the wiring is pre-programming and doesn't change,
> > > connections change, but can not be made.
> >
> > Until the shared memory libs was written, you couldn't change any Eu
> > program either. You couldn't execute sequences as if they were Eu code,
> > except in David Cuny's interpreter. We still can't launch threads, but
> > parallel processing without real threads may very well be an enormous
> > blessing
> > in disguise. The other problem is longevity, winOSs haveto get rebooted
> > occasionally, and getting the program to remember what it knows and how it
> > knows, and what to do with what it knows, and the last state of the "brain",
> > will be a pain, because (for example) we cannot look up the list of vars
> > (names and contents) like Mirc or Lua can do.
> >
> > Kat
> >
>
> Wow. You sound almost like a Dredge fan.
>
> Seriously, the "simulated" neural net is mostly data structures.
> The hard-coded stuff will be the initial state of the data strucutres
> and the Eu code itself which actually maniplulates the data structures.
If all the Eu code is hard coded, then how will it learn how to do anything
new?
Kat
10. Re: AI Project
On 0, Kat <kat at kogeijin.com> wrote:
> > Seriously, the "simulated" neural net is mostly data structures.
> > The hard-coded stuff will be the initial state of the data strucutres
> > and the Eu code itself which actually maniplulates the data structures.
>
> If all the Eu code is hard coded, then how will it learn how to do anything
> new?
>
> Kat
>
The information isnt in the Eu code.
Its stored in some other mechanism, like a neural net (which is stored as
a bunch of sequences).
The Eu code itself just manipulates the neural net on predefined
values (say whats good, or bad, basicly control the weighing of inputs
and
converting outputs, etc.) THAT has to be predefined. Everything which
doesnt,
does not have to be in Eu code, but just stored as part of the nerual net
(which can be a sequence,a nd thus can be modified).
jbrown
--
