In this homework you will use Terry Regier's program to learn
spatial relation terms. 

Steps
-----

a) copy the program space.tar to your directory. 
(type "cp /home/ee/cs182/space.tar ~/.")

b) type "tar -xf space.tar" in your directory.

c) type "cd NonPunctate" to get to the static scene learning directory.

d) Answer questions by training and testing on the data. 

Training and Testing
-------------------

To view all scenes in the training set for English "above", type

do see above

To train on English "above", type

do train above

To test on English "above" type

do test above.

To train/test/view_scenes for other words, replace "above" with the
appropriate word.

[Note: ignore the dark square]

*********************************************************************

The information below is NOT REQD. for your assignment, but
may be useful if you want to play with the program on your own.

In case you are interested:

Other Command line options include:

	-a <fname>:	Take fname as the architecture file
	-c <fname>:	Learn concept fname
	-o:		Use old weights (from an earlier training run)
	-t:		Want to train (not just show results)
	-T <fname>:	Use fname as the test set
	-m <num>:	Mode switch threshold for quickprop (if high, e.g.
			100, program basically uses ordinary backprop.
			See Fahlman's quickprop tech report for details.)
	-E <num>:	Epsilon (learning rate).
	-s <num>:	Start off with this many epochs.
	-[ <num>:	Freeze incoming weights for nodes <num> through ...
	-] <num>:	      ... <num>.
	-z:		Stop training when zero bits wrong

.........................................................................

Stray tidbits:

   Some of the programs in here call each other:

	scene calls get_theta once the scene file has been written,
		so that get_theta can compute the proximal TR orientation
		for the scene.

	train calls allscene when done, so as to display all the 
		training set and the learned values for each of the scenes.

.........................................................................