Merge pull request #18 from jxx123/random_init_bg

Randomize init BG for the gym env

simglucose/envs/simglucose_gym_env.py

@@ -36,7 +36,7 @@ def __init__(self, patient_name=None, reward_fun=None, seed=None):
         self.patient_name = patient_name
         self.reward_fun = reward_fun
         self.np_random, _ = seeding.np_random(seed=seed)
-        self.env, _, _ = self._create_env_from_random_state()
+        self.env, _, _, _ = self._create_env_from_random_state()
 
     def _step(self, action):
         # This gym only controls basal insulin
@@ -46,30 +46,31 @@ def _step(self, action):
         return self.env.step(act, reward_fun=self.reward_fun)
 
     def _reset(self):
-        self.env, _, _ = self._create_env_from_random_state()
+        self.env, _, _, _ = self._create_env_from_random_state()
         obs, _, _, _ = self.env.reset()
         return obs
 
     def _seed(self, seed=None):
         self.np_random, seed1 = seeding.np_random(seed=seed)
-        self.env, seed2, seed3 = self._create_env_from_random_state()
-        return [seed1, seed2, seed3]
+        self.env, seed2, seed3, seed4 = self._create_env_from_random_state()
+        return [seed1, seed2, seed3, seed4]
 
     def _create_env_from_random_state(self):
         # Derive a random seed. This gets passed as a uint, but gets
         # checked as an int elsewhere, so we need to keep it below
         # 2**31.
         seed2 = seeding.hash_seed(self.np_random.randint(0, 1000)) % 2**31
         seed3 = seeding.hash_seed(seed2 + 1) % 2**31
+        seed4 = seeding.hash_seed(seed3 + 1) % 2**31
 
         hour = self.np_random.randint(low=0.0, high=24.0)
         start_time = datetime(2018, 1, 1, hour, 0, 0)
-        patient = T1DPatient.withName(self.patient_name)
+        patient = T1DPatient.withName(self.patient_name, random_init_bg=True, seed=seed4)
         sensor = CGMSensor.withName(self.SENSOR_HARDWARE, seed=seed2)
         scenario = RandomScenario(start_time=start_time, seed=seed3)
         pump = InsulinPump.withName(self.INSULIN_PUMP_HARDWARE)
         env = _T1DSimEnv(patient, sensor, pump, scenario)
-        return env, seed2, seed3
+        return env, seed2, seed3, seed4
 
     def _render(self, mode='human', close=False):
         self.env.render(close=close)

simglucose/patient/t1dpatient.py

@@ -19,7 +19,7 @@ class T1DPatient(Patient):
     SAMPLE_TIME = 1  # min
     EAT_RATE = 5    # g/min CHO
 
-    def __init__(self, params, init_state=None, t0=0):
+    def __init__(self, params, init_state=None, random_init_bg=False, seed=None, t0=0):
         '''
         T1DPatient constructor.
         Inputs:
@@ -30,9 +30,9 @@ def __init__(self, params, init_state=None, t0=0):
             - t0: simulation start time, it is 0 by default
         '''
         self._params = params
-        if init_state is None:
-            init_state = self._params.iloc[2:15]
-        self.init_state = init_state
+        self._init_state = init_state
+        self.random_init_bg = random_init_bg
+        self._seed = seed
         self.t0 = t0
         self.reset()
 
@@ -229,10 +229,38 @@ def _announce_meal(self, meal):
             to_eat = 0
         return to_eat
 
+    @property
+    def seed(self):
+        return self._seed
+
+    @seed.setter
+    def seed(self, seed):
+        self._seed = seed
+        self.reset()
+
     def reset(self):
         '''
         Reset the patient state to default intial state
         '''
+        if self._init_state is None:
+            self.init_state = self._params.iloc[2:15]
+        else:
+            self.init_state = self._init_state
+
+        self.random_state = np.random.RandomState(self.seed)
+        if self.random_init_bg:
+            # Only randomize glucose related states, x4, x5, and x13
+            mean = [1.0 * self.init_state[3], 
+                    1.0 * self.init_state[4], 
+                    1.0 * self.init_state[12]]
+            cov = np.diag([0.1 * self.init_state[3], 
+                           0.1 * self.init_state[4], 
+                           0.1 * self.init_state[12]]) 
+            bg_init = self.random_state.multivariate_normal(mean, cov)
+            self.init_state[3] = 1.0 * bg_init[0]
+            self.init_state[4] = 1.0 * bg_init[1]
+            self.init_state[12] = 1.0 * bg_init[2]
+
         self._last_Qsto = self.init_state[0] + self.init_state[1]
         self._last_foodtaken = 0
         self.name = self._params.Name