Parameters
-
simulator (can be 50chain, blackjack, pong, tetris, wumpus, blocks or logistics)
-
transfer (can be 0 or 1)
-
number_of_iterations
-
batch_size
-
loss (can be LS, LAD or Huber)
-
trees (number of trees)
Example run: FVI(simulator="blocks",loss="Huber",number_of_iterations=10)
Summary of changes
-
Discretized features in propositional domains. (Before propositional baselines were using continuous features)
-
Plotted new graphs in graphs folder
Notes
-
Results may differ due to high variance.
-
For stable results and to perform as per theoretical expectation best to increase initial model computation iterations in compute transfer model function and/or increase batch size and fix the policy during comparison in the execute random action function present in all simulators
contact: kxr150330@utdallas.edu
This is still in development. So, please wait for a while until it is pefected. Go ahead and test the version that is there, I'd love to learn about more issues that the code might have