fix refactor bug

utils/reward_func.py

@@ -0,0 +1,21 @@
+import math
+import torch
+import numpy as np
+from collections import deque
+from utils import core
+from utils.core import custom_reward, custom_reward_traj
+
+
+def composite_reward(args, state=None, reward=None):
+    MAX_GLUCOSE = 600
+    if reward == None:
+        reward = custom_reward([state])
+    x_max, x_min = 0, custom_reward([MAX_GLUCOSE]) #get_IS_Rew(MAX_GLUCOSE, 4) # custom_reward([MAX_GLUCOSE])
+    reward = ((reward - x_min) / (x_max - x_min))
+    if state <= 40:
+        reward = -15
+    elif state >= MAX_GLUCOSE:
+        reward = 0
+    else:
+        reward = reward
+    return reward

utils/statespace.py

@@ -0,0 +1,82 @@
+import math
+import torch
+import numpy as np
+from collections import deque
+from utils import core
+
+class StateSpace:
+    def __init__(self, args):
+        self.feature_history = args.feature_history
+        self.glucose = deque(self.feature_history*[0], self.feature_history)
+        self.insulin = deque(self.feature_history*[0], self.feature_history)
+        self.glucose_max = args.glucose_max
+        self.glucose_min = args.glucose_min
+        self.insulin_max = args.insulin_max
+        self.insulin_min = args.insulin_min
+        self.t_meal = args.t_meal
+        self.use_carb_announcement = args.use_carb_announcement
+        self.mealAnnounce = args.use_meal_announcement
+        self.todAnnounce = args.use_tod_announcement
+
+        if not self.mealAnnounce and not self.todAnnounce:  # only ins and glucose
+            self.state = np.stack((self.glucose, self.insulin), axis=-1).astype(np.float32)
+        elif self.todAnnounce:
+            self.meal_announcement_arr = deque(self.feature_history * [0], self.feature_history)
+            self.tod_announcement_arr = deque(self.feature_history * [0], self.feature_history)
+            self.hc_iob_20 = deque(self.feature_history * [0], self.feature_history)
+            self.hc_iob_60 = deque(self.feature_history * [0], self.feature_history)
+            self.hc_iob_120 = deque(self.feature_history * [0], self.feature_history)
+            self.state = np.stack((self.glucose, self.insulin, self.meal_announcement_arr, self.tod_announcement_arr,
+                                   self.hc_iob_20, self.hc_iob_60, self.hc_iob_120), axis=-1).astype(np.float32)
+        elif self.use_carb_announcement:
+            self.meal_announcement_arr = deque(self.feature_history * [0], self.feature_history)
+            self.carb_announcement_arr = deque(self.feature_history * [0], self.feature_history)
+            self.state = np.stack((self.glucose, self.insulin, self.meal_announcement_arr, self.carb_announcement_arr), axis=-1).astype(np.float32)
+        else:
+            self.meal_announcement_arr = deque(self.feature_history * [0], self.feature_history)
+            # self.meal_type_arr = deque(self.feature_history * [0], self.feature_history)
+            self.state = np.stack((self.glucose, self.insulin, self.meal_announcement_arr), axis=-1).astype(np.float32)
+
+    def update(self, cgm=0, ins=0, meal=0, hour=0, meal_type=0, carbs=0):
+        cgm = core.linear_scaling(x=cgm, x_min=self.glucose_min, x_max=self.glucose_max)
+        ins = core.linear_scaling(x=ins, x_min=self.insulin_min, x_max=self.insulin_max)
+        hour = core.linear_scaling(x=hour, x_min=0, x_max=312)  # hour is given 0-23
+        t_to_meal = core.linear_scaling(x=meal, x_min=0, x_max=self.t_meal)  #self.t_meal * -2
+        snack, main_meal = 0, 0
+        if meal_type == 0.3:
+            snack = 1
+        elif meal_type == 1:
+            main_meal = 1
+
+        meal_type = core.linear_scaling(x=meal_type, x_min=0, x_max=1)
+        carbs = core.linear_scaling(x=carbs, x_min=0, x_max=120)
+        self.glucose.append(cgm)  # self.glucose.appendleft(cgm)
+        self.insulin.append(ins)
+
+        # handcrafted features
+        ins_20 = sum(self.state[-4:, 1])  # last 20 minreward
+        ins_60 = sum(self.state[-12:, 1])  # last 60 min
+        ins_120 = sum(self.state[-24:, 1])  # last 120 min
+
+        if not self.mealAnnounce and not self.todAnnounce:
+            self.state = np.stack((self.glucose, self.insulin), axis=-1).astype(np.float32)
+        elif self.todAnnounce:
+            self.meal_announcement_arr.append(t_to_meal)
+            self.tod_announcement_arr.append(hour)
+            self.hc_iob_20.append(ins_20)
+            self.hc_iob_60.append(ins_60)
+            self.hc_iob_120.append(ins_120)
+            self.state = np.stack((self.glucose, self.insulin, self.meal_announcement_arr, self.tod_announcement_arr,
+                                   self.hc_iob_20, self.hc_iob_60, self.hc_iob_120), axis=-1).astype(np.float32)
+        elif self.use_carb_announcement:
+            self.meal_announcement_arr.append(t_to_meal)
+            self.carb_announcement_arr.append(carbs)
+            self.state = np.stack((self.glucose, self.insulin, self.meal_announcement_arr, self.carb_announcement_arr), axis=-1).astype(np.float32)
+        else:
+            self.meal_announcement_arr.append(t_to_meal)
+            # self.meal_type_arr.append(meal_type)
+            self.state = np.stack((self.glucose, self.insulin, self.meal_announcement_arr), axis=-1).astype(np.float32)
+
+        #handcraft_features = [cgm, ins, ins_20, ins_60, ins_120, hour, t_to_meal, snack, main_meal]
+        handcraft_features = [hour]
+        return self.state, handcraft_features