DTMDP with Minimum Tree Reinitialization

With previous decision tree results, I alway reinitialized a tree to the zero computational cost tree for each epsilon value. In these results, I store, for each epsilon value, the tree that yields the minimum expected cost. Then for the the next epsilon value, I use the minimum decision tree as the initializer. This yielded better results than reinitializing with the zero computational cost tree for higher ranges of epsilon. There were instances where the best value decreased with increasing iterations (hightlighted in red). Yellow highlightling shows expected value improvement. This was not the case when I reinitialized using the zero cost tree. It is clear that the initialization procedure has an impact on subsequent tree generation.