tldr;
A greedy search policy that at each step evaluates the reward that is produced by every possible action and selects the one with greatest reward, or with some probability ε will choose to select an action randomly.
Authors: Facebook AI Research
Results:
Publication:
CompilerGym version: 0.1.4
Open source? Yes, MIT licensed. Source Code.
Did you modify the CompilerGym source code? No.
What parameters does the approach have? Probability of selecting a random action ε.
What range of values were considered for the above parameters? ε=0 (greedy).
Is the policy deterministic? Yes if ε=0, else no.
At each step the agent selects either a greedy policy or an exploration policy. If greedy, every action is evaluated and the action with the greatest reward is selected. If exploration, an action is selected randomly. The episode terminates when the maximum reward attainable by any action is <= 0.
| Hardware Specification | |
|---|---|
| OS | Ubuntu 20.04 |
| CPU | Intel Xeon Gold 6230 CPU @ 2.10GHz (80× core) |
| Memory | 754.5 GiB |
$ python e_greedy.py --n=1 --epsilon=0 --results_logfile=results_e0.csv