3/11/2011  DL

This is similar to test1 except that it tries to use a single output whose level
is the answer rather than multiple outputs. This allows the ANN to interpolate between
training points to predict solutions in areas where it has no data.

Because the functions used by the ANN have a range 0-1, the inputs and outputs are
scaled in this test by 1/10. Furthermore, the inputs and output are shifted by 1 (before
and after the scaling respectively) so that the range of inputs is 0.1-0.4 rather than
0-3. This is more in line with the 0.0-1.0 range of the internal functions. The shifting
avoids input values close to 0 which can be problematic.

The ANN has 2 inputs which take the values of the 2 numbers to be multiplied. There are 3 hdden
layers with 8, 10, and 10 neurons respectively. There is a single output representing the number
that is the result.

To run the test and plot the results do this:
make
./make_train_file
./train_network
./test_network
root -q -b output_vs_actual.C