RL experiment

Codes/Julia/Part-3/reinforcement-learning.jl

@@ -0,0 +1,40 @@
+############################
+#= REINFORCEMENT LEARNING =#
+############################
+
+using ReinforcementLearning
+using Flux: Descent
+
+## Define the environment
+env = RandomWalk1D()
+
+## Instantiate the agent
+agent = Agent(
+    policy = QBasedPolicy(
+        learner = TDLearner(
+            approximator = TabularQApproximator(
+                n_state = 11,
+                n_action = 2,
+                init = 0.0,
+                opt = Descent(0.1) # Learning rate
+            ),
+            method = :SARSA,
+            γ = 0.99
+        ),
+        explorer = EpsilonGreedyExplorer(0.1),
+    ),
+    trajectory = VectorSARTTrajectory(),
+)
+
+## Run the experiment
+hook = TotalRewardPerEpisode()
+run(agent, env, StopAfterEpisode(10_000), hook)
+
+## Print rewards
+println("Total reward per episode:")
+println(hook.rewards)
+
+## Print `Q-table``
+q_table = agent.policy.learner.approximator.table
+println("\nLearned Q-table:")
+println(q_table)