/Preferential-Temporal-Difference-Learning

Code to reproduce results of the paper - Preferential Temporal Difference Learning.

Primary LanguagePython

Preferential Temporal Difference Learning

Tabular

To run the tabular experiments (Figure 2) run the following command:

cd Tabular
./run_tabular.sh

Linear

Run the following commands to reproduce Figure 3 and A.2.

cd LinearFA
./run_task1.sh
./run_task2.sh

The next two settings requires pytorch to reproduce the figures.

Semi-Linear

First run the feature net by executing the following command:

cd Semi_LinearFA
./run_MC.sh

Running the following commands will reproduce results presented in Figure 4 and A.8 by loading the trained feature net models.

./run_task1.sh
./run_task2.sh

Non-Linear

Running the following commands will generate Figures 5 and 6 from the paper. The first two commands correspond to the forward-view results and the next two commands correspond to the backward-view results.

Comment data generation part inside each script file if your Data folder is not empty

cd NonLinearFA
./run_task1.sh
./run_task2.sh
./run_task1_traces.sh
./run_task2_traces.sh

All figures are saved in plots directory of the respective settings. For example, the tabular figures are saved in Tabular/plots/ directory.

Results are saved in results_ directories after and you can delete them after generating the plot.