Data for DTMDP for Various Tree Generation Iterations

This shows the best value achieved by the DTMDP is reached only after one iteration of the algorithm. The highlighted cells in the table indicate an improvement over the previous trial. I tried a range of iterations: 1,5,10, and 20. It appears that the DTMDP solution does not improve after the 20th iteration.

The value of the DTMDP solution is not monotonically decreasing with the number of iterations, but bounces around. The values achieved from bouncing around appear to be within a bounded set, meaning that the various DTMDP averages for a given epsilon take on only a finite set of values.

The entire 2^n graph value states were used to generate new decision trees. The graph configuration used to generate these results was the Simple Graph Problem with 8 arcs and 6 nodes.

My next step will be to determine the mechanics of tree generation and whether there is some manner to force the solution to converge in a monotonic manner.