Today on reddit, someone asked about freely available machine learning API's.
Before the list gets buried, I'm duplicating the contents of that thread here for future exploration:
Weka - Java based ML API
http://www.cs.waikato.ac.nz/ml/weka/
Toolkit for Advanced Disrimnative Modeling
http://tadm.sf.net/
Mallet - Java based ML API
http://mallet.cs.umass.edu/
WekaUT - An extension of Weka that adds clustering
http://www.cs.utexas.edu/users/ml/risc/code/
LibSVM - SVM API
http://www.csie.ntu.edu.tw/~cjlin/libsvm/
SVMlight - A C API for svm
http://svmlight.joachims.org/
C++ API for Neural Networks
http://github.com/bayerj/arac
Torch5 - A Matlab-like ML environment
http://torch5.sourceforge.net/
R - Open source statistical package that can be used for ML
http://cran.r-project.org/web/views/MachineLearning.html
pyML - Python API for ML
http://pyml.sourceforge.net/
Rapidminer - Open Source Data Mining Tool
http://rapid-i.com/wiki/index.php?title=Main_Page
Orange - An Open Source Data Mining Tool (Python and GUI based)
http://www.ailab.si/orange/
Glue - Open Source API for reinforcement learning (Can be used with multiple languages simultaneously)
http://glue.rl-community.org/wiki/Main_Page
Vowpal Rabiit - Learning API from Yahoo Research
http://hunch.net/~vw/
Showing posts with label AI. Show all posts
Showing posts with label AI. Show all posts
Wednesday, October 14, 2009
Tuesday, October 6, 2009
Natural Language Generation
While going over some nbc's, I stumbled across the AI area of NLP. But, while observing where the areas of nbc and NLP meet, I've found a new obsession: NLG. NLG is an acronym for natural language generation. Natural language generation is text created by a computer program that appears to be human-like in readability.
I first heard about this topic in detail in my AI class in grad school at University of San Francisco. My professor, Dr. Brooks, had mentioned that researchers had been trying for years to create programs that could generate narratives for computer games. I even recall seeing on some news aggregator that someone had successfully won a writing contest with a story written by a NLG system.
At the time I was taking my AI course, I remember working for a horrible boss. Who made all of us who he saw everyday and interacted with on a continuous basis, write weekly reports. I remember wanting to write a Perl or Python script that would do this for me. I made some attempts but it was hard to get any realistic variance. It was essentially an overglorified mad lib, where the program only filled in the blanks.
I was looking for something more natural and human like.
In NLG, one takes data and has generation rules that result in text that feels as if a human wrote it. Surprisingly, if one does a search on NLG, it is a relatively new area of research. Perhaps the best introduction to this topic is on Wikipedia. From the Wikipedia area, you will find yourself on the Bateman and Zock list of Natural Language Generators(http://www.fb10.uni-bremen.de/anglistik/langpro/NLG-table/NLG-table-root.htm)
At the moment, the state of the art appears to be based upon Java and Lisp languages. Since I work in embedded systems where speed and small footprint are key, I'm intersted in implementations that are in C and can scale. I've noticed that most of the NLP and NLG systems I found do not have a database backend. This surprises me since use of a database would allow for scaling and more consistant performacne as the dataset grows.
I think I'll be experimenting with NLG to see if I can make a program that will generate an email that asks a user for info based upon an email inquiry.
I first heard about this topic in detail in my AI class in grad school at University of San Francisco. My professor, Dr. Brooks, had mentioned that researchers had been trying for years to create programs that could generate narratives for computer games. I even recall seeing on some news aggregator that someone had successfully won a writing contest with a story written by a NLG system.
At the time I was taking my AI course, I remember working for a horrible boss. Who made all of us who he saw everyday and interacted with on a continuous basis, write weekly reports. I remember wanting to write a Perl or Python script that would do this for me. I made some attempts but it was hard to get any realistic variance. It was essentially an overglorified mad lib, where the program only filled in the blanks.
I was looking for something more natural and human like.
In NLG, one takes data and has generation rules that result in text that feels as if a human wrote it. Surprisingly, if one does a search on NLG, it is a relatively new area of research. Perhaps the best introduction to this topic is on Wikipedia. From the Wikipedia area, you will find yourself on the Bateman and Zock list of Natural Language Generators(http://www.fb10.uni-bremen.de/anglistik/langpro/NLG-table/NLG-table-root.htm)
At the moment, the state of the art appears to be based upon Java and Lisp languages. Since I work in embedded systems where speed and small footprint are key, I'm intersted in implementations that are in C and can scale. I've noticed that most of the NLP and NLG systems I found do not have a database backend. This surprises me since use of a database would allow for scaling and more consistant performacne as the dataset grows.
I think I'll be experimenting with NLG to see if I can make a program that will generate an email that asks a user for info based upon an email inquiry.
Friday, September 25, 2009
Grammars
The purpose of this entry is to describe the 4 type of grammars that can be used to classify a language, and the means used to classify a language as one of the four types of grammrs.
From a linguistics and NLP standpoint, languages can be classified by 4 possible grammar types. From the most expressive description to the least expressive description, a language can be described by a grammar known as type 0, type 1, type 2, or type 3. Each of the different grammars has a common name. A given grammar can sometimes describe other grammars. A type 0 grammar can describe type 1, type2, and type 3 grammars. A type 1 grammar can describe type 2 and type 3. A type 2 grammar can describe a type 3 grammar. A type 3 grammar cannot describe any other type of grammar.
For the four types of grammars, each are composed of rules known as productions, which have the general form of w1 -> w2 . Each production rule generates a sequence of terminal tokens and non-terminals. Non-terminals are production rules that go by the lhs symbol, w1 .
A recursively enumerable grammar is also known as a type 0 grammar. A type 0 grammar has no restrictions on its production rules. Context-sensitive grammar is a type 1 grammar. A type 1 grammar is restricted to productions where the number of symbols on the rhs is equal to or greater than the number of symbols on the lhs. Context-free grammar is a type 2 grammar. A non-terminal in a type 2 grammar can be replaced by its rhs. In comparison, a non-terminal in a type 1 grammar can only be replaced if there is a production that matches the symbols on the rhs with an equivalent lhs. A regular grammar is a type 3 grammar. A regular grammar is also known as a regular expression, which is used by Perl, Python, and grep when searching on strings. A production of a regular grammar has a restricted expression. The lhs is a non-terminal. The rhs is a terminal, which is optionally followed by a non-terminal.
Grammars are also more formally known as phrase structure grammars.
G = phrase structure grammar as a set
G = (V,T,S,P)
V is a vocabulary, a set of tokens and non-tokens(non-terminals)
T is a subset of V. T is the set of terminal tokens
S is a start symbol/token that is a member of V
P is a set of production rules
N is V - T, set of non-terminal symbols/tokens
Types of grammars and restrictions on their productions, w1 -> w2
0 no restrictions
1 length(w1) <= length(w2), w2=lambda
2 w1 = A, where A is non-terminal symbol
3 w1=A and w2=aB or w2=a, where A is an element of N, B is an element of N, and a is an element of T, or S->lambda
From a linguistics and NLP standpoint, languages can be classified by 4 possible grammar types. From the most expressive description to the least expressive description, a language can be described by a grammar known as type 0, type 1, type 2, or type 3. Each of the different grammars has a common name. A given grammar can sometimes describe other grammars. A type 0 grammar can describe type 1, type2, and type 3 grammars. A type 1 grammar can describe type 2 and type 3. A type 2 grammar can describe a type 3 grammar. A type 3 grammar cannot describe any other type of grammar.
For the four types of grammars, each are composed of rules known as productions, which have the general form of w1 -> w2 . Each production rule generates a sequence of terminal tokens and non-terminals. Non-terminals are production rules that go by the lhs symbol, w1 .
A recursively enumerable grammar is also known as a type 0 grammar. A type 0 grammar has no restrictions on its production rules. Context-sensitive grammar is a type 1 grammar. A type 1 grammar is restricted to productions where the number of symbols on the rhs is equal to or greater than the number of symbols on the lhs. Context-free grammar is a type 2 grammar. A non-terminal in a type 2 grammar can be replaced by its rhs. In comparison, a non-terminal in a type 1 grammar can only be replaced if there is a production that matches the symbols on the rhs with an equivalent lhs. A regular grammar is a type 3 grammar. A regular grammar is also known as a regular expression, which is used by Perl, Python, and grep when searching on strings. A production of a regular grammar has a restricted expression. The lhs is a non-terminal. The rhs is a terminal, which is optionally followed by a non-terminal.
Grammars are also more formally known as phrase structure grammars.
G = phrase structure grammar as a set
G = (V,T,S,P)
V is a vocabulary, a set of tokens and non-tokens(non-terminals)
T is a subset of V. T is the set of terminal tokens
S is a start symbol/token that is a member of V
P is a set of production rules
N is V - T, set of non-terminal symbols/tokens
Types of grammars and restrictions on their productions, w1 -> w2
0 no restrictions
1 length(w1) <= length(w2), w2=lambda
2 w1 = A, where A is non-terminal symbol
3 w1=A and w2=aB or w2=a, where A is an element of N, B is an element of N, and a is an element of T, or S->lambda
Sunday, September 20, 2009
Tracing the boot up sequence of TAEB part 2
This entry goes into detail of the top of the main loop of the taeb script, line 81-98.
taeb,lines 81-98 is the top-level of the taeb main loop. The main loop is composed of 2 main parts. The first item in the main loop is the actual operational statement. The later item, which is actually multiple statements, are only executed if taeb has been told via the command line option, --loop, to re-execute itself after it has completed playing a nethack session.
Let's review these 2 portions by first going over the later item, since it is only executed when specified via the command-line.
taeb, line88: only reinitialize/restart taeb if local variable $loop is not equal to 0. This will happen when --loop is specified via the command-line.
taeb, line90: reset all taeb variables and states
taeb, lines 92-96: Sleep for 5 seconds before starting a new nethack game
The eval statement, which is the first statement ecountered at the start of this loop, does all the work.
taeb, line 83: sets INT handler to print message to screen and prevent starting a new game by setting $loop to 0.
taeb, line 84: execute a taeb session
taeb, line 85: reports the results of a taeb session
taeb, line84, is TAEB->play;The method play is sent to the TAEB object, which is defined in TAEB.pm.
TAEB.pm, lines 742-749: Inside this loop, each iteration is a step. At the end of each step, results are store of the step. The last step of the taeb session will print its results to the screen.
Inside each step, the following methods are called in order:
redraw
display_topline
human_input
full_input
handle_XXXX called
redraw is invoked from TAEB::Display::Curses. redraw is used to repaint the entire nethack screen at the start of a step. display_topline is also invoked from TAEB::Display::Curses, too. display_topline displays any messages received from the last step in the current step. human_input is defined inside TAEB.pm. human_input is used to get keyboard input if ai allows human control and a key is pressed. Under normal operational circumstances, human_input will not capture anything.
full_input is a wraper method that is used to capture nethack screen data and load it into the taeb database. At the start of full_input's operation, the screen scraper is reset to take new info and the publisher is turned off. The publisher is resumed after scraping and processing of the scrape is finished. After the publisher is off, then next instruction performed is process_input, which reads any input from a previous action or user and sends it to the vt, virtual terminal. Afterwords, the screen is scraped and loaded into the Taeb database.
The next step performed is that dungeon and senses variables are updated. After the update of these items, the publisher is renabled. This completes the phase of a step where the percepts are captured.
taeb,lines 81-98 is the top-level of the taeb main loop. The main loop is composed of 2 main parts. The first item in the main loop is the actual operational statement. The later item, which is actually multiple statements, are only executed if taeb has been told via the command line option, --loop, to re-execute itself after it has completed playing a nethack session.
Let's review these 2 portions by first going over the later item, since it is only executed when specified via the command-line.
taeb, line88: only reinitialize/restart taeb if local variable $loop is not equal to 0. This will happen when --loop is specified via the command-line.
taeb, line90: reset all taeb variables and states
taeb, lines 92-96: Sleep for 5 seconds before starting a new nethack game
The eval statement, which is the first statement ecountered at the start of this loop, does all the work.
taeb, line 83: sets INT handler to print message to screen and prevent starting a new game by setting $loop to 0.
taeb, line 84: execute a taeb session
taeb, line 85: reports the results of a taeb session
taeb, line84, is TAEB->play;The method play is sent to the TAEB object, which is defined in TAEB.pm.
TAEB.pm, lines 742-749: Inside this loop, each iteration is a step. At the end of each step, results are store of the step. The last step of the taeb session will print its results to the screen.
Inside each step, the following methods are called in order:
redraw
display_topline
human_input
full_input
handle_XXXX called
redraw is invoked from TAEB::Display::Curses. redraw is used to repaint the entire nethack screen at the start of a step. display_topline is also invoked from TAEB::Display::Curses, too. display_topline displays any messages received from the last step in the current step. human_input is defined inside TAEB.pm. human_input is used to get keyboard input if ai allows human control and a key is pressed. Under normal operational circumstances, human_input will not capture anything.
full_input is a wraper method that is used to capture nethack screen data and load it into the taeb database. At the start of full_input's operation, the screen scraper is reset to take new info and the publisher is turned off. The publisher is resumed after scraping and processing of the scrape is finished. After the publisher is off, then next instruction performed is process_input, which reads any input from a previous action or user and sends it to the vt, virtual terminal. Afterwords, the screen is scraped and loaded into the Taeb database.
The next step performed is that dungeon and senses variables are updated. After the update of these items, the publisher is renabled. This completes the phase of a step where the percepts are captured.
Saturday, September 19, 2009
Tracing the boot up sequence of TAEB part 1
The purpose of this entry is to describe in plain talk how Taeb starts up and executes it 'main' loop. By understanding the 'main' loop of the taeb framework, this will enable creation of Taeb agents.
taeb, line1: enables text file to be interpreted as a Perl script
taeb, line2: no questionable Perl constructs allowed
taeb, line3: ? literally says to add to @INC the the library using the literal 'lib'
taeb, line4: Use Perl library Getopt::Long to process command-line options
taeb, lines 7-20: defintion of print_usage subroutine which shows the command-line options that can be used to configure Taeb operation
taeb, line22: local variable used to control whether or not taeb will just stop when it receives an interrupt or reset and begin a new execution. See while-loop at line 81.
taeb, line23: local list that stores command-line specified options for taeb. This local list is then transferred to Taeb's configureation. See lines 24, 28, 43.
taeb, line24: local hash that is initialzed to store the hard reference to the variable $loop and the list @config_overrides.
taeb, lines 25-39: Code used to alternate operation of Taeb via command-line options. Also, used to display the options that can be specified to Taeb.
taeb, line41: Specify that TAEB.pm must be used and found.
taeb, line43: Change Taeb configuration from specified command-line options
taeb, lines 45-47: Add to Taeb configruation that no ai should be used if command-line option specified.
taeb, lines 49-77: handlers assigned to signals TSTP, CONT, TERM, USR1, and USR2 signals.
taeb, line79: Flush after every write
taeb, lines 81-98: Top of the main loop of Taeb.
taeb, line1: enables text file to be interpreted as a Perl script
taeb, line2: no questionable Perl constructs allowed
taeb, line3: ? literally says to add to @INC the the library using the literal 'lib'
taeb, line4: Use Perl library Getopt::Long to process command-line options
taeb, lines 7-20: defintion of print_usage subroutine which shows the command-line options that can be used to configure Taeb operation
taeb, line22: local variable used to control whether or not taeb will just stop when it receives an interrupt or reset and begin a new execution. See while-loop at line 81.
taeb, line23: local list that stores command-line specified options for taeb. This local list is then transferred to Taeb's configureation. See lines 24, 28, 43.
taeb, line24: local hash that is initialzed to store the hard reference to the variable $loop and the list @config_overrides.
taeb, lines 25-39: Code used to alternate operation of Taeb via command-line options. Also, used to display the options that can be specified to Taeb.
taeb, line41: Specify that TAEB.pm must be used and found.
taeb, line43: Change Taeb configuration from specified command-line options
taeb, lines 45-47: Add to Taeb configruation that no ai should be used if command-line option specified.
taeb, lines 49-77: handlers assigned to signals TSTP, CONT, TERM, USR1, and USR2 signals.
taeb, line79: Flush after every write
taeb, lines 81-98: Top of the main loop of Taeb.
Saturday, September 12, 2009
Demo AI with TAEB and try_explore part 1
TAEB ships with an AI known as Demo. one can use it as a reference for creating your own AI with TAEB. At its most simplistic level, the creation of a TAEB AI invovles the following steps:
1. Create a Perl module that inherits from TAEB::AI
2. Create a method called next_action that returns a an object of type TAEB::Action
I will be describing how next_action in Demo moves around the the Nethack dungeon via next_action sending the try_explore method. In TAEB's Demo AI, one of the actions that it can execute is defined by the try_explore method. From the current position in the Nethack dungeon, TAEB will use try_explore to find the next best tile to reach amount it's 8 nearest neighbors.
try_explore itself is merely a wrapper. The work is done first_match, which takes as an argument a type of the destination tile. In this case, the destination tile type is 'unexplored'. Tile types are defined in lib/TAEB/Util.pm from lines 73-91. first_match is part of TAEB::World::Path package. The purpose of first_match is to identify the type of data structure that will be used to search for the next tile position. Typically, for the purposes to try_explore this will be undef. After determining the type of data structure for searching, first_match calls _dijskstra, which is an implementation of the Dijkstra path finding algorithm.
1. Create a Perl module that inherits from TAEB::AI
2. Create a method called next_action that returns a an object of type TAEB::Action
I will be describing how next_action in Demo moves around the the Nethack dungeon via next_action sending the try_explore method. In TAEB's Demo AI, one of the actions that it can execute is defined by the try_explore method. From the current position in the Nethack dungeon, TAEB will use try_explore to find the next best tile to reach amount it's 8 nearest neighbors.
try_explore itself is merely a wrapper. The work is done first_match, which takes as an argument a type of the destination tile. In this case, the destination tile type is 'unexplored'. Tile types are defined in lib/TAEB/Util.pm from lines 73-91. first_match is part of TAEB::World::Path package. The purpose of first_match is to identify the type of data structure that will be used to search for the next tile position. Typically, for the purposes to try_explore this will be undef. After determining the type of data structure for searching, first_match calls _dijskstra, which is an implementation of the Dijkstra path finding algorithm.
Saturday, September 5, 2009
Alternate means of running a Taeb AI agent
The easiest way to use an AI agent other than Demo in Taeb is to install the new agent into your Taeb installation. If you do this, then you only need to have a config.yml in your .taeb directory that specifies this new agent.
But, if for some reason, you don't want the agent's source code to be installed into your main installation, there is another way. As before, you will still need a config.yml in your .taeb file. If you have followed the software conventions for Taeb agents, cd into your agent's lib directory;The lib directory only contains the directory 'TAEB'. Inside that directory, lib, which is considered the 'top' of your agent's code, run 'taeb'. This will grab the agent defined under 'taeb' as the agent to run.
One can get the same results if one sets the full path of PERL5LIB to the 'lib' directory of your agent. This method has the advantage that only a config.yml is needed. No need to run taeb from the specific lib directory of your agent.
But, if for some reason, you don't want the agent's source code to be installed into your main installation, there is another way. As before, you will still need a config.yml in your .taeb file. If you have followed the software conventions for Taeb agents, cd into your agent's lib directory;The lib directory only contains the directory 'TAEB'. Inside that directory, lib, which is considered the 'top' of your agent's code, run 'taeb'. This will grab the agent defined under 'taeb' as the agent to run.
One can get the same results if one sets the full path of PERL5LIB to the 'lib' directory of your agent. This method has the advantage that only a config.yml is needed. No need to run taeb from the specific lib directory of your agent.
Friday, September 4, 2009
TAEB keyboard commands
The keyboard commands that can be used to interact with a TAEB-based agent are defined in TAEB/lib/TAEB.pm
p - pause agent
d - change draw mode
i - show inventory
\cP - show old message
\cX - Senses
e - equipment menu
I - item spoiler data
M - monster spoiler data
\e - user input
r - refresh screen
\cr - refresh screen
q - save and exit
Q - controlled quit and exit
p - pause agent
d - change draw mode
i - show inventory
\cP - show old message
\cX - Senses
e - equipment menu
I - item spoiler data
M - monster spoiler data
\e - user input
r - refresh screen
\cr - refresh screen
q - save and exit
Q - controlled quit and exit
How to stop a TAEB AI properly
shift+Q
If you try some other key sequence, such as ctrl-c, it will just stop, but when you re-run taeb again it will pick up where you left off. Most of the time I just want to start over.
Many key commands I think are located in lib/TAEB.pm in the TAEB source.
If you try some other key sequence, such as ctrl-c, it will just stop, but when you re-run taeb again it will pick up where you left off. Most of the time I just want to start over.
Many key commands I think are located in lib/TAEB.pm in the TAEB source.
Structure of TAEB AI Behavioral Part 3
For the personality described the the Perl module Explorer.pm, each time the method next_action is called from the base class TAEB::AI::Behavioral::Personality, the member list is scanned for the behavior with the highest urgency.
The member list this is scanned is called prioritized_behaviors, which is an array of strings that are the humand-readable name of a possible behavior of the agent. The elements of prioritized_behaviors are(listed in the order of scanning from top to bottom) is: FixHunger, Heal, FixStatus, Defend, AttackSpell, BuffSelf, Kite, Melee, ProjectFiles, Vault, Shop, Carrion, GetItems, Equip, Identify, DipForExcalibur, Wish.
When the next_action is called it first performs any post-behaviors by sending the done method to complete any steps from the current behavior, if defined. The behavior Luckstone appears to be the only behavior with an implementation of the done method. After calling done, next_action will scan the entire prioritized_behaviors array, looking fof the first behavior with the highest urgency. If there are multiple behaviors with the same urgency and the urgency is the largest, then the first one found is the most urgent behavior. As next_action goes through each behavior in the prioritized_behaviors list, it executes each behavior's prepare method. The prepare method of a TAEB::AI::Behavior instance performs setup in preperation for a behavior to be chosen as the most urgent next behavior. For example, for the Explore.pm behavior, the prepare method performs a search of the nethack maze from its current position. When next_action completes its iterations over the contents of the prioritized_behaviors, all behaviors have had their prepare methods executed, but only one behavior is the most urgent and will be converted into a nethack command, TAEB::AI::Action subclass.
The member list this is scanned is called prioritized_behaviors, which is an array of strings that are the humand-readable name of a possible behavior of the agent. The elements of prioritized_behaviors are(listed in the order of scanning from top to bottom) is: FixHunger, Heal, FixStatus, Defend, AttackSpell, BuffSelf, Kite, Melee, ProjectFiles, Vault, Shop, Carrion, GetItems, Equip, Identify, DipForExcalibur, Wish.
When the next_action is called it first performs any post-behaviors by sending the done method to complete any steps from the current behavior, if defined. The behavior Luckstone appears to be the only behavior with an implementation of the done method. After calling done, next_action will scan the entire prioritized_behaviors array, looking fof the first behavior with the highest urgency. If there are multiple behaviors with the same urgency and the urgency is the largest, then the first one found is the most urgent behavior. As next_action goes through each behavior in the prioritized_behaviors list, it executes each behavior's prepare method. The prepare method of a TAEB::AI::Behavior instance performs setup in preperation for a behavior to be chosen as the most urgent next behavior. For example, for the Explore.pm behavior, the prepare method performs a search of the nethack maze from its current position. When next_action completes its iterations over the contents of the prioritized_behaviors, all behaviors have had their prepare methods executed, but only one behavior is the most urgent and will be converted into a nethack command, TAEB::AI::Action subclass.
Thursday, September 3, 2009
Structure of the Taeb AI Behavioral Part 2
The Behavioral AI has 4 possible personalities which are defined by the Perl pm's Explorer, ScoreWhore, Bathophobe, and Descender. Explorer.pm is the base personality. ScoreWhore, Bathophobe, and Descender are all based upon Explorer.pm.
When Explorer.pm is instantiated by the Taeb framework, a hash, %behaviors, is created. This has uses the string names of actions as keys;These keys are mapped to instances of a TAEB::AI::Behavioral::Behavior::actionStringName. Explorer.pm contains an array of strings called prioritzed_behaviors.
Each time next_action is called, the work is done in Behavioral.pm. When next_action is called, the first task accomplished is that the criticalness of each action, which we will now call a behavior, will be calculated. The criticalness of each behavior is a property which is called urgency, which is either a string or numeric value. The string values of urgency are critical, important, normal, unimportant, fallback, and none. The corresponding numerical values are 50, 40, 30, 20, 10, and 0.
The method next_action calculates the urgency for each of the behaviors in the prioritized_behaviors. The urgency is calculated by the find_urgency method. For each behavior, find_urgency examines the current state of the agent with respect to the environment and sets the urgency member value. As the urgency is calculated, next_action is always looking for the behavior with the largest urgency. If multiple items have the same urgency value, the behavior that is in the prioritized_behavior first is the behavior that will be selected for generating a nethack action. Sometimes, no urgency is calculated. In the case of Explore.pm, the prepare method merely executes the Explore behavior each time Explore is examined during the next_action method.
When Explorer.pm is instantiated by the Taeb framework, a hash, %behaviors, is created. This has uses the string names of actions as keys;These keys are mapped to instances of a TAEB::AI::Behavioral::Behavior::actionStringName. Explorer.pm contains an array of strings called prioritzed_behaviors.
Each time next_action is called, the work is done in Behavioral.pm. When next_action is called, the first task accomplished is that the criticalness of each action, which we will now call a behavior, will be calculated. The criticalness of each behavior is a property which is called urgency, which is either a string or numeric value. The string values of urgency are critical, important, normal, unimportant, fallback, and none. The corresponding numerical values are 50, 40, 30, 20, 10, and 0.
The method next_action calculates the urgency for each of the behaviors in the prioritized_behaviors. The urgency is calculated by the find_urgency method. For each behavior, find_urgency examines the current state of the agent with respect to the environment and sets the urgency member value. As the urgency is calculated, next_action is always looking for the behavior with the largest urgency. If multiple items have the same urgency value, the behavior that is in the prioritized_behavior first is the behavior that will be selected for generating a nethack action. Sometimes, no urgency is calculated. In the case of Explore.pm, the prepare method merely executes the Explore behavior each time Explore is examined during the next_action method.
Wednesday, September 2, 2009
Structure of the Taeb AI Behavioral Part 1
This is the start of a series of entries on the Taeb AI agent known as Behavioral.
Taeb is an AI agent written in Perl5. Within the Taeb framework, the Behaviorial agent would be found within lib/TAEB/AI module.
Overall, Behavioral is composed of 5 modules:
lib/TAEB/AI
lib/TAEB/AI/Behavorial
lib/TAEB/AI/Behavioral/Behavior
lib/TAEB/AI/Behavioral/Meta
lib/TAEB/AI/Behavioral/Personality
lib/TAEB/AI contains the Behavioral module and the Perl module Behavioral.pm. Behavioral.pm Behavioral.pm will cause a user to use a specific subclass of the Behavioral class.
The Behavioral module contains the Behavior, Meta, and Personality modules. The Behavior module contains the various actions an agent can perform. Meta defines the urgency type, which is used to choose a particular action. Personality contains the various types of Behavioral agents. In addition, the Perl modules Behavior.pm, Personality.pm, and ThreatEvaluation.pm live here. Every agent can accomplish some sort of action. Behavior.pm is a base class that defines the common attributes of an action. Personality.pm is the base class of the different types of Behavioral agents. ThreatEvaluation.pm is a data object used to determine the relationship between resource consumption and attacking a monster.
In the Personality module, there are 4 different types of Behavioral agents: Bathophobe, Descender, Explorer, and ScoreWhore. In the config.yml file, one specifies one of these Behavioral agent types.
Taeb is an AI agent written in Perl5. Within the Taeb framework, the Behaviorial agent would be found within lib/TAEB/AI module.
Overall, Behavioral is composed of 5 modules:
lib/TAEB/AI
lib/TAEB/AI/Behavorial
lib/TAEB/AI/Behavioral/Behavior
lib/TAEB/AI/Behavioral/Meta
lib/TAEB/AI/Behavioral/Personality
lib/TAEB/AI contains the Behavioral module and the Perl module Behavioral.pm. Behavioral.pm Behavioral.pm will cause a user to use a specific subclass of the Behavioral class.
The Behavioral module contains the Behavior, Meta, and Personality modules. The Behavior module contains the various actions an agent can perform. Meta defines the urgency type, which is used to choose a particular action. Personality contains the various types of Behavioral agents. In addition, the Perl modules Behavior.pm, Personality.pm, and ThreatEvaluation.pm live here. Every agent can accomplish some sort of action. Behavior.pm is a base class that defines the common attributes of an action. Personality.pm is the base class of the different types of Behavioral agents. ThreatEvaluation.pm is a data object used to determine the relationship between resource consumption and attacking a monster.
In the Personality module, there are 4 different types of Behavioral agents: Bathophobe, Descender, Explorer, and ScoreWhore. In the config.yml file, one specifies one of these Behavioral agent types.
Sunday, August 30, 2009
Configuring TAEB at launch to use a different AI(and other settings)
I've been playing around with Taeb and Nethack for a few days now. I'm really glad I ignored this game during college. I have a few friends who didn't and didn't finish.
I gave myself a crash course review of Perl. I mainly did this so I could understand the code for Demo.pm, which is the reference agent. I wanted to try to change agents, but I mistakenly thought this could be done with taeb's --config option.
In order to change the AI of Taeb, one must place a yml file into $HOME/.taeb. The yml file must be renamed to config.yml
I gave myself a crash course review of Perl. I mainly did this so I could understand the code for Demo.pm, which is the reference agent. I wanted to try to change agents, but I mistakenly thought this could be done with taeb's --config option.
In order to change the AI of Taeb, one must place a yml file into $HOME/.taeb. The yml file must be renamed to config.yml
Saturday, August 29, 2009
Setup of TAEB on openSUSE 11.1
TAEB stands for Tactical Amulet Extraction Bot. It is a framework written Perl. This framework enables one to make a bot that can play nethack. I'm interested in TAEB because I want to try to reduce its runtime footprint and experiment with bots that use search based AI and rules based AI. In the recent 2009 Mario AI contest, preliminary results showed that A* search implementations were beating the rules based implementations. I found this suprising because I thought rules based AI would be more flexible and faster.
I intend to do these same types of experiments with TAEB.
First step was to get nethack installed. nethack was available at www.nethack.org. OpenSUSE 11.1's current repositories alos had nethack available, so I just installed it with the OpenSUSE software installer.
I downloaded the latest TAEB using git. git clone git://github.com/sartak/TAEB.git
I discovered that if I attempted to install TAEB using sudo: sudo perl Makefile.PL
For some reason, the Perl function can_run can't find my installation of nethack. This was odd. It was necessary to create a symbolic link /usr/bin/nethack to /usr/games/nethack.
After making the symbolic link to the nethack executable, there were a number of warnings about additional Perl modules not found on the system:
I went back to eliminate each of these warnings. Unless specified otherwise, I used the cpan executable to eliminate these warnings.
Note, I found that Yaml::Scyk had to be installed but Makefile.PL did not detect and list that it was needed; When I ran taeb the first time, it complaed that it couldn't find the file YAML/Syck.pm.
After YAML::Syck was installed, I rebuilt and reinstalled. Seems to be working!
I intend to do these same types of experiments with TAEB.
First step was to get nethack installed. nethack was available at www.nethack.org. OpenSUSE 11.1's current repositories alos had nethack available, so I just installed it with the OpenSUSE software installer.
I downloaded the latest TAEB using git. git clone git://github.com/sartak/TAEB.git
I discovered that if I attempted to install TAEB using sudo: sudo perl Makefile.PL
For some reason, the Perl function can_run can't find my installation of nethack. This was odd. It was necessary to create a symbolic link /usr/bin/nethack to /usr/games/nethack.
After making the symbolic link to the nethack executable, there were a number of warnings about additional Perl modules not found on the system:
I went back to eliminate each of these warnings. Unless specified otherwise, I used the cpan executable to eliminate these warnings.
Note, I found that Yaml::Scyk had to be installed but Makefile.PL did not detect and list that it was needed; When I ran taeb the first time, it complaed that it couldn't find the file YAML/Syck.pm.
After YAML::Syck was installed, I rebuilt and reinstalled. Seems to be working!
Wednesday, August 26, 2009
Naive Bayes Classifiers in SpamAssassin
Spam classifiers like SpamAssassin are broadly used to split email into ham and spam. How well would SpamAssassin's nbc perform if there were more than 2 categories. I have an idea for applying nbc of a spam filter for sorting emails which will be split into more than 2 categories.
As a start of this investigation, I've decided to start with some OSS Naive Bayes Classifer based spam filters. I'm starting with SpamAssassin. For the purposes of this experiment, I will be using SpamAssassin as a command-line tool.
spamassassin is a Perl front-end that one uses to classify an email, which is in a text file. One email per file.
sa-learn is a tool in the SpamAssassin suite that trains the nbc.
sa-learn --ham /path/to/directory/containing/ham loads the nbc with ham.
sa-leanr --spam /path/to/directory/containing/spam loads the nbc with spam.
I've only acquired a corpus of ham and spam of a few thousand emails. For what I need, I would like to have a corpus of up to a million documents which could be split into about 9 categories. I'm looking for a large corpus.
I've also noted with nbc's that process text, I've notice that there appears to be no restriction on the email size. In comparison to nbc's used with images, it is required that the images in the image corpus all be the same size. I wonder if this is really necessary. I will have to check that out with the face recognizer work currently in progress in my OpenCV project.
As a start of this investigation, I've decided to start with some OSS Naive Bayes Classifer based spam filters. I'm starting with SpamAssassin. For the purposes of this experiment, I will be using SpamAssassin as a command-line tool.
spamassassin is a Perl front-end that one uses to classify an email, which is in a text file. One email per file.
sa-learn is a tool in the SpamAssassin suite that trains the nbc.
sa-learn --ham /path/to/directory/containing/ham loads the nbc with ham.
sa-leanr --spam /path/to/directory/containing/spam loads the nbc with spam.
I've only acquired a corpus of ham and spam of a few thousand emails. For what I need, I would like to have a corpus of up to a million documents which could be split into about 9 categories. I'm looking for a large corpus.
I've also noted with nbc's that process text, I've notice that there appears to be no restriction on the email size. In comparison to nbc's used with images, it is required that the images in the image corpus all be the same size. I wonder if this is really necessary. I will have to check that out with the face recognizer work currently in progress in my OpenCV project.
Sunday, August 23, 2009
Naive Bayes for text classification
Another article on Naive Bayes classification aka nbc? Why does the web need another article on this. Yes, there are many. Just use your favorite browser and search on "naive bayes". You'll get hundreds of url's. However, once you start digging deep into them, you will find 2 annoying trends. They are just reprinting the Naive Bayes definition. No specifics on how the Naive Bayes definition was used to classify a type of data.
Originally, I had leared about Naive Bayes in my AI class, which was taught by Dr. Christopher Brooks, at University of San Francisco back in 2005. Since then, the computer that held my information died and I've been unable to retrieve my nbc program. After a long time, I finally found good info on Naive Bayes and how to use it toward classifying text. First, I'll give the basic definitions of Bayes and Naive Bayes classification, which I'm simply resummarizing from the text book "Artificial Intelligence" by Russel and Norvig. Then, I'll talk about specifically applying Naive Bayes for text classification; This information I found in slides for a course called Comp221. The slides were written by the courses TA, Zhang Kai.
Naive Bayes
Like the regular Bayes algorithm, Naive Bayes simplifies the calculation of a conditional probability. It further simplifies the calculation of conditional probabilities by assuming that the effects are independent. Even though this may not be actually true, it has been found that this assumption yields acceptable behavior.
P(Class|Effects) = P(Class) * P(Effect1|Class).....*P(Effectn|Class)
Supervised Naive Bayes for Text Classification
The definition of Naive Bayes is easy to understand, but is lacking in any of the details that one must use to make a real application. I will fill in the details here(Thanks Dr. Brooks and Zhang Kai!)
(1) Start with a corpus and calculate P(Ci)
A corpus is a collection of data that will be used to train the Naive Bayes classiifer. This should be large number of items. The total number items should be around 1000. The corpus should be split into different classes, where each class occurs in the percentage one thinks the actual documents occur in real life. Ci is a class in C. P(Ci) = nc / ni, where nc is the number of corpus documents that correspond to Ci. ni is the total number of documents in the corpus.
(2)For each class, calculate the P(word|class)
For each class, there will be a collection of words that are associated with that class. One must calculate the probablity that a given word will occur in a particular class.
ni = number of total words in documents in Ci
wi = word associated with Ci
wij = number of times wi occurs in all Ci documents
P(wi|Ci) = wij / ni
For each class, if a word only occurs in the class Ci, this is considered a conditional probability of 'zero'. For an conditional probability that is a 'zero', assign it a value eta/ni, where eta/ni is some tunable value.
For each class, choose the top word frequencies as the words used to classify a document. Ideally, each word would occur in all Ci in C.
(3) After (1) and (2) have been performed, the nbc has been trained. d is a new document that is unclassified.
Take document d and find all the words that occur in the training corpus.
For each Ci, calculate P(Ci|Effects); For each word, wi, calculate the P(wi|Ci) wrt to the document d.
The largest P(Ci|Effects) is the matched class Ci for document d.
Originally, I had leared about Naive Bayes in my AI class, which was taught by Dr. Christopher Brooks, at University of San Francisco back in 2005. Since then, the computer that held my information died and I've been unable to retrieve my nbc program. After a long time, I finally found good info on Naive Bayes and how to use it toward classifying text. First, I'll give the basic definitions of Bayes and Naive Bayes classification, which I'm simply resummarizing from the text book "Artificial Intelligence" by Russel and Norvig. Then, I'll talk about specifically applying Naive Bayes for text classification; This information I found in slides for a course called Comp221. The slides were written by the courses TA, Zhang Kai.
Naive Bayes
Like the regular Bayes algorithm, Naive Bayes simplifies the calculation of a conditional probability. It further simplifies the calculation of conditional probabilities by assuming that the effects are independent. Even though this may not be actually true, it has been found that this assumption yields acceptable behavior.
P(Class|Effects) = P(Class) * P(Effect1|Class).....*P(Effectn|Class)
Supervised Naive Bayes for Text Classification
The definition of Naive Bayes is easy to understand, but is lacking in any of the details that one must use to make a real application. I will fill in the details here(Thanks Dr. Brooks and Zhang Kai!)
(1) Start with a corpus and calculate P(Ci)
A corpus is a collection of data that will be used to train the Naive Bayes classiifer. This should be large number of items. The total number items should be around 1000. The corpus should be split into different classes, where each class occurs in the percentage one thinks the actual documents occur in real life. Ci is a class in C. P(Ci) = nc / ni, where nc is the number of corpus documents that correspond to Ci. ni is the total number of documents in the corpus.
(2)For each class, calculate the P(word|class)
For each class, there will be a collection of words that are associated with that class. One must calculate the probablity that a given word will occur in a particular class.
ni = number of total words in documents in Ci
wi = word associated with Ci
wij = number of times wi occurs in all Ci documents
P(wi|Ci) = wij / ni
For each class, if a word only occurs in the class Ci, this is considered a conditional probability of 'zero'. For an conditional probability that is a 'zero', assign it a value eta/ni, where eta/ni is some tunable value.
For each class, choose the top word frequencies as the words used to classify a document. Ideally, each word would occur in all Ci in C.
(3) After (1) and (2) have been performed, the nbc has been trained. d is a new document that is unclassified.
Take document d and find all the words that occur in the training corpus.
For each Ci, calculate P(Ci|Effects); For each word, wi, calculate the P(wi|Ci) wrt to the document d.
The largest P(Ci|Effects) is the matched class Ci for document d.
Subscribe to:
Posts (Atom)