[src/experiment] Add DDDQN-70-90 experiment

indyfree · indyfree · commit dea13234c822 · 2019-04-21T15:31:17.000+02:00
diff --git a/src/evsim/envs/fleet_env.py b/src/evsim/envs/fleet_env.py
@@ -57,8 +57,8 @@ def seed(self, seed=None):
 
     def step(self, action):
         # Transform "flat" action back to tuple
-        rb = (action // 11) / 10
-        ri = (action % 11) / 10
+        rb = (action // 11) / 20
+        ri = (action % 11) / 20
 
         balance, done = self.sim.step(risk=(rb, ri), minutes=15)
         reward = balance - self.curr_balance
diff --git a/src/evsim/experiments/DDDQN-70-90.py b/src/evsim/experiments/DDDQN-70-90.py
@@ -0,0 +1,26 @@
+#!/usr/bin/env python
+
+import gym
+
+from evsim.experiments import setup_logger
+from evsim.rl import DDQN
+
+name = "DDDQN-70-90"
+episodes = 2
+episode_steps = 65334
+
+setup_logger("sim-{}".format(name), write=False)
+env = gym.make("evsim-v0")
+env.imbalance_costs(5000)
+env.prediction_accuracy((80, 95))
+
+dqqn = DDQN(
+    env,
+    name,
+    memory_limit=(episodes * episode_steps),
+    nb_eps=(1.5 * episode_steps),
+    nb_warmup=1000,
+)
+dqqn.run(episodes * episode_steps)
+
+dqqn.test()
