Readme for the Agent Building and Learning Environment
Version 1.1b September 29, 2000
The Agent Building and Learning Environment (ABLE) is a
toolkit from the IBM T.J. Watson Research Center for developing
hybrid intelligent software agents and agent applications in Java.
Readme topics are:
Overview
Licensing
Release Contents
Requirements
Installation and
Running
Known Problems
and Limitations
Support
The Agent Building and Learning Environment (ABLE) is a Java-based
toolkit for developing and deploying hybrid intelligent agents
and agent applications. ABLE provides a set of reusable JavaBean
components, called AbleBeans, along with several flexible
interconnection methods for combining those components to create
software agents. AbleBeans implement data access, filtering and
transformation, learning, and reasoning capabilities. Function-specific
AbleAgents are provided for classification, clustering,
prediction, and genetic search. Application-specific agents can
be constructed using one or more of these AbleBeans. AbleAgents
are situated in their environment through the use of sensors and
effectors, which provide a generic mechanism for linking them to
Java applications. A GUI-based interactive development
environment, the Able Agent Editor, is provided to assist in the
construction of AbleAgents using AbleBean components.
In the ABLE framework, an agent is an autonomous software
component. It could be running on its own thread of control or
could be called synchronously by another agent or process either
through a direct method call or by sending an event. By combining
one or more AbleBeans, agents can be extremely lightweight (e.g.
a few lines of Java code) or can be relatively heavy weight,
using multiple forms of inferencing (e.g. fuzzy rule systems,
forward and backward chaining) and learning (e.g. neural
classification, prediction, and clustering).
AbleBeans provided include:
- AbleImport
- reads data from flat text files
- AbleFilter
- filters, transforms, and scales data using translate
template specifications
- AbleExport
- writes data to flat text files
- Back Propagation
- implements enhanced back propagation algorithm used for
classification and prediction
- Self-Organizing Map
- supports batch updates, Gaussian neighborhood function
- Boolean inference engine
- processes if-then rules using forward and backward
chaining
- Fuzzy inference engine
- processes if-then rules containing fuzzy sets and
supports multi-step forward chaining
- Naive Bayes
- learns a probabalistic model for classification
- Radial Basis Function
- uses radial basis functions to adjust weights in a single
hidden layer neural network
- Temporal Difference Learning
- uses reinforcement learning for time series forecasting;
gradient descent is used to adjust network weights
-
AbleAgents provided include:
- Genetic search agent
- manipulates a population of genetic objects which may
include AbleBeans
- Neural classifier agent
- uses back propagation to classify data
- Neural clustering agent
- uses self-organizing maps to segment data
- Neural prediction agent
- uses back propagation to build regression models
Please read the License
Agreement and agree to the terms and conditions before using
this software.
Release 1.1b September 29, 2000
Major changes are:
- JDK 1.1.8 is no longer supported; minimum JDK version is
1.2.2 with 1.3 recommended. Agents serialized with prior
Able releases will not be compatible.
- Rule languages support XML, which requires developers to
add the xerces.jar file to the CLASSPATH setting.
- The Able package is now split into code and documentation
to make downloads faster.
- Learning beans have been expanded with the addition of AbleNaiveBayes,
AbleRadialBasisFunctionNet and AbleTemporalDifferenceLearning,
and a new network graphic inspector.
- Neural agents have autotrain capability, which switches
modes between train and test as well switching data
inputs between import beans.
- A description of how to extend
Able by providing your own beans is added to the user
reference, and is available separately in PDF
format. It refers to a new sample bean AbleFileWatcher
provided for monitoring file changes.
- The Able Data Model has been externalized in a new
package com.ibm.able.data which will allow people to
create new data types for inferencing/filtering/importing/exporting
by extending the newly exposed data types. A description
in PDF
format is available.
- A new definition file format is in use. The previous
version is read and converted to the new format; the old
design did not support multiple target outputs and did
not share well between algorithms. Import beans can now
read data in sequential or random order.
- Miscellaneous Agent Editor enhancements and cleanup such
as displaying bean state in the tree view, better
handling of the icons on the canvas, able to define a
directory from which to read user-provided jars, user-provided
beans can specify their own tab on the editor bean
palette, consistent look and feel throughout.
- Removed examples/neural/* file from ableall.jar and
ableexamples.jar which caused "class not found
errors" with some JDK versions on Windows.
- Workaround provided for JDK bug which prevented help
browser from opening for JDK 1.2+ users.
Release 1.0a April 28, 2000
This is the first release of the ABLE package.
Platform: JDK 1.2.2 or later.
Computer: Win 95/98/NT, UNIX, OS/2
ABLE Installation Procedures
- Download the able1_1b.zip file or the able1_1b.tar.gz
file as appropriate. To reduce download times, help and
documentation is provided in a separate file, able1_1bDocs,
which you can download separately.
- Unzip (or untar) the file(s) to a specified drive. If you
chose to download both the executable and the help and
documentation files, use the same root directory for each.
The .../able1_1b directory tree will be created
and populated.
Script files to start the Able Agent Editor are in the
.../able1_1b/bin directory.
- Documentation can be found in the .../able1_1b/index.html
file.
Running the Able Agent Editor
- Change to the .../able1_1b/bin directory.
- Run the runnit script appropriate for your system.
This is an alpha level release and not all function and
documentation is complete. In particular:
- Serialized agents from previous releases will not be
compatible.
- Filter editor documentation is not complete.
- Inspector help is not provided.
- JDK 1.2 has documented a bug that may cause panels to
contain ghost images of other panels (Sun bug 4265726).
We have implemented a workaround which may not fix this
problem in all cases, and has a side effect that in some
instances causes some screen flicker.
See the 'Troubleshooting FAQ' section in the FAQ if you
encounter problems installing or using ABLE.
We are interested in hearing of any problems you encounter
while using this software. Please send comments and questions to ableinfo@us.ibm.com or post
them in the Able discussion forum at http://www.alphaworks.ibm.com/tech/able.