Running learning-by-demonstration (as opposed to
interactive learning) with pre-recorded solutions
See also Running Interactive Learning in batch-mode
You can train SimStudent with a bunch of pre-recorded solutions (saved as
BRD files, the files generated by Behavior Recorder). In this case,
SimStudent learns steps in the BRD files as if they are demonstrated through
the Student Interface.
You can also test production rules generated by SimStudent with a set of
test problems, also stored as the BRD files. Namely, SimStudent opens a BRD
file and, for each step in the BRD, reports production rules that are
applicable and if they generate
the same step as recorded in the BRD.
||Save BRD files
You can either use Behavior Recorder to create BRD files
manually by just demonstrating solutions once for each, or write a
piece of software to automatically generate BRD files (which is
especially useful when you have a dataset containing student-tutor
||Make a shell script
All what you need is to call a Java class file for the Student
Interface with some command line arguments.
The first command
line argument must be "ssBatchMode". You may also specify following command line
arguments (see <url> for detailed descriptions for the command
- ssProblemSet -- a list of
training "files" to be used as the pre-recorded demonstrations.
- ssTestSet -- a list of test
"files" to test production rules learned by SimStudent
Both for ssProblemSet and ssTestSet, a "file" can be
either a name of a BRD file or a directory name. If the
"file" represents a directory, then all the BRD files in the
directory will be iteratively used for training.
A sample shell script is available <here>.
Once you run SimStudent, you have a few things that you may want
to pay attention:
||A text file contains the production rules that have
been learned most recently. If you are interested in how
production rules had been learned, then take a look at
the pr-age folder, which is described below.
||A folder (directory) where the past
productionRules.pr files are archived. There are two
types of archive files shown below:
contains a total of yyy production rules that
are learned in zzz demonstrated steps.
which shows the production rules learned after the
xxx-th training problem is completed with the
total of yyy production rules learned in zzz steps.
January 24, 2010