dupa

Author Topic: Learn from text file limitation  (Read 3317 times)

Kaane

  • Newbie
  • *
  • Posts: 9
    • View Profile
Learn from text file limitation
« on: June 07, 2004, 09:52:34 pm »
Greetz to All!

We have put extensive time into trying to give UltraHal a very in-depth technical education in physics to determine the functional use of ultrahal as an AI agent in unattended problem solving, but we're running into a few bumps in the road... first off, we have a huge problem with an appearent size limitation in the text file that UH can learn from, on all of our boxes it seems to be in the vicinity of 65k. Is there any way of overcoming this usable text size in a *.txt file?

Second, using the uncensored brain and some indepth data provided to UH in text files, when questions are asked that can easily be answered by data in the text files provided, about 25% of the time it answers questions correctly... the other 75% of the time it responds to questions with a question like "How loud can you fart?" or "Can we hump each other?". We have tried this with just the default brain also but when questioned, it tends to answer with other nonsense questions the majority of the time. How can we place greater focus on having UH answer the question at hand, especially when we *know* the question can be answered correctly if UH refers to the data previously provided to it in text files?

As it stands, I would swear UH is definately female.

Help??

Regards,

Kaane
 

NT Canuck

  • Newbie
  • *
  • Posts: 38
    • View Profile
    • http://ntcanuck.com
Learn from text file limitation
« Reply #1 on: June 08, 2004, 09:28:59 am »
quote:
Originally posted by Kaane

Greetz to All!
[]
of 65k. Is there any way of overcoming this usable text size in a *.txt file?
[]
especially when we *know* the question can be answered correctly
Kaane



65 kb per file, max 2mb is retained.
no sdk. no sources. afaik

-

sounds like meta-search bot called by uhal would do,
that is somewhat similar to aibush program that displays
info from databases in a browser.  if you want ai to parse
then needs some targeting (i haven't had chance to confirm)
along lines of attached...but that bot is limited mem also.

'Seek and ye shall find'
NT Canuck
http://ntcanuck.com
_________

Targeting - what is it? and why it might be useful
======================
This release did some work to bring back a relatively unknown feature of the Program D code called "targeting." It is useful for analyzing how people talk to your robot and to create new programming to support better conversations.

To invoke Targeting, just type the following in the command line:
/targeting.sh <<then press return>>

A graphical interface will appear and will begin to analyze log files. You will soon see some interesting info.

There are three main windows which you can select by clicking on tabs at the bottom of the screen, they are:
Targets
Inputs
Categories

Targets Window - This shows that a chat user typed a particular "input." Then, it shows what that input "matched" in the AIML logic. Next, it shows what you may want to create for a "new category" in AIML. You can modify the various fields to your liking and then modify the "reply" field and then click "save category." Once it is saved, it is put into a file located in "targets/targets.aiml" where you can later open up in a text editor and move into your robots brain.

Here is an example:
INPUT:      what is dos
MATCHED:   WHAT IS *
NEW CATEGORY:   WHAT IS DOS
REPLY:      What about it? DOS stands for Desparate Orthodontic
      Services...no it really stands for Disk Operating
      System. It was the first Operating system for the
      IBM PC. It was created after the Apple had already
      been released.

Inputs Window - The Inputs Window lists all the different inputs users gave in list form along with the patterns that they matched. You can double click on any of the items in the list to be taken to the "Targets Window" where you can edit and save.

Categories Window - The Categories Window shows you how many times various patterns have been matched. The number of times is the number of "activations." This list view is useful for determining which parts of your AIML are getting hit the most often and are probably most deserving of your tender loving care. Here to you may double click on any of the items in the list to go to the "Targeting Window" for editing.

_________

# Alicebot Program D Targeting Tool Configuration
#------------------------------------------------------------
# Alicebot Program D
# Copyright (C) 1995-2001, A.L.I.C.E. AI Foundation
#
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU General Public License
# as published by the Free Software Foundation; either version 2
# of the License, or (at your option) any later version.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307,
# USA.

# -----------------------------------------------------------------------------
# GENERAL NOTES
# -----------------------------------------------------------------------------

# This is a properties file just for the Targeting Tool.  Some properties
# specified here are also specified in the server.properties, but we use this
# file just to focus on the properties that are needed for the Targeting Tool.
# You can specify the use of this (or another) by specifying it as a command-line
# argument to org.alicebot.core.targeting.TargetingTool.


# From where targeting data should be read
programd.targeting.data.path=./targets/targets.xml

# Delay period when checking targets data from Targeting Tool (milliseconds)
programd.targeting.tool.reload-timer=5000

# Where generated AIML should be written
programd.targeting.tool.aiml.path=./targets/targets.aiml

# Whether to show targets whose match patterns were complete but whose <that> patterns were not.
programd.targeting.tool.include-incomplete-thats=false

# Whether to show targets whose match patterns were complete but whose <topic> patterns were not.
programd.targeting.tool.include-incomplete-topics=false

# Preferred encoding for writing generated AIML from Targeting Tool (default: UTF-8)
#programd.targeting.tool.aiml.encoding=ISO-8859-5

# -----------------------------------------------------------------------------
# CONSOLE/TRACE CONFIGURATION
# -----------------------------------------------------------------------------

# Show information on console (true/false)
programd.console=true

# Show developer info messages on console (true/false)
programd.console.developer=false

# Show message type flags on console (true/false)
programd.console.message-flags=false

# The date-time format to use on the console
# * See http://java.sun.com/j2se/1.3/docs/api/java/text/SimpleDateFormat.html
#   for formatting codes.
# * Setting the value to blank means no timestamp will be displayed.
programd.console.timestamp-format=H:mm:ss
« Last Edit: June 08, 2004, 09:38:38 am by NT Canuck »
 

Bill819

  • Hero Member
  • *****
  • Posts: 1483
    • View Profile
Learn from text file limitation
« Reply #2 on: June 08, 2004, 11:44:46 am »
quote:
Originally posted by Kaane

Greetz to All!

We have put extensive time into trying to give UltraHal a very in-depth technical education in physics to determine the functional use of ultrahal as an AI agent in unattended problem solving, but we're running into a few bumps in the road... first off, we have a huge problem with an appearent size limitation in the text file that UH can learn from, on all of our boxes it seems to be in the vicinity of 65k. Is there any way of overcoming this usable text size in a *.txt file?

Second, using the uncensored brain and some indepth data provided to UH in text files, when questions are asked that can easily be answered by data in the text files provided, about 25% of the time it answers questions correctly... the other 75% of the time it responds to questions with a question like "How loud can you fart?" or "Can we hump each other?". We have tried this with just the default brain also but when questioned, it tends to answer with other nonsense questions the majority of the time. How can we place greater focus on having UH answer the question at hand, especially when we *know* the question can be answered correctly if UH refers to the data previously provided to it in text files?

As it stands, I would swear UH is definately female.

Help??

Regards,

Kaane


Look for the member ONTHECUTTINGEDGE here. He has his own web page and I believe that he has already solved your problem for physics and several other topics as well.
Bill