An IRC conversation simulator (chatbot) utilising Markov chains, Hebbian networks and information theory.
Triplie is an AI bot based on 2nd up to 5th order Markov model. It uses an SQLite database for storage.
Triplie learns by creating
To respond to a user, triplie extracts keywords from the user's text, finds their most appropriate associated keywords in the Hebbian association network, and generates replies that contain the associated keywords using multiple breadth-first-search Markov chains algorithm.
For more information on installing and configuring read below
You can join the project's IRC channel too: #triplie on irc.freenode.net
Download and install node.js for your system. Its recommended to build node from source. If you don't do that, make sure that npm is also installed alongside with node and that the node binary is called "node"
Then from a terminal run:
npm install -g triplie
This will install the
triplie command on your system.
Configure the bot as explained below before running!
If running the bot for the first time and its not configured, you should create a new directory and run:
triplie config.yaml --init
to create the initial config file
config.yaml is already pre-filled with some default for your bot. You will want to change some of these settings.
The configuration file is really well commented. Open it and edit it according to the instructions contained inside. Once you run the bot however, the instructions will disappear the moment you change a setting by giving a command to the bot.
After you edited the config file, to run the bot use the command:
If called with the argument
--feed triplie will receive data from stdin,
parse it using a regular expression then feed the database.
cat log.txt | triplie config.yaml --feed --regex '(?<year>\d+)-(?<month>\d+)-(?<day>)T(?<hour>\d+):(?<minute>\d+):(?<second>\d+)Z\s+(?<nick>.+):\s+(?<text>.+)'
will work for a
log.txt that has lines in the format:
2013-04-04T13:15:00Z someuser: I wrote some text
The syntax is XRegExp and uses named groups. See the XRegExp readme for more info
Currently, supported named captures are:
cat log.txt | triplie config.yaml --feed --regex '(?<timestamp>\d+) (?<text>.+)
log.txt containing lines in the format:
1234567890 example text here
All captures except text are optional - the time is optional and if left out the feeder will generate reasonable "fake" timestamps.
cat log.txt | triplie config.yaml --feed --regex '(?<text>.+)'
List of triplie's commands (assuming "!" is the cmdchar)
!join #channel - causes the bot to join and remember the channel
!part #channel - part and forget channel
!reload - causes reload of the bot code, useful for development
!set path value - set a config setting to the specified value. Examples
!set ai.sleep.1 10 - Set the upper sleep limit to 10 seconds !set ai.sleep [2,3] - Set both sleep limits. Value musn't contain space.
!get path - get the config value at the specified path
!db stats - triplie will output database statistics
!cmd will return results via private notice
!!cmd returns results via public message
See LICENCE and AUTHORS (if present)