def risk_diff(BG_last_hour):
if len(BG_last_hour) < 2:
return 0
else:
_, _, risk_current = risk_index([BG_last_hour[-1]], 1)
_, _, risk_prev = risk_index([BG_last_hour[-2]], 1)
return risk_prev - risk_current
def step(self, action, reward_fun=neg_risk):
'''
action is a namedtuple with keys: basal, bolus
'''
CHO = 0.0
insulin = 0.0
BG = 0.0
CGM = 0.0
# This loop runs for n minutes in patients life
# n is determined by the sample time, which is in turn defined by
# the quality of sensor and the interval between each measurement
for _ in range(int(self.sample_time)):
# Compute moving average as the sample measurements
tmp_CHO, tmp_insulin, tmp_BG, tmp_CGM = self.mini_step(action)
CHO += tmp_CHO / self.sample_time
insulin += tmp_insulin / self.sample_time
BG += tmp_BG / self.sample_time
CGM += tmp_CGM / self.sample_time
# Compute risk index
horizon = 1
LBGI, HBGI, risk = risk_index([BG], horizon)
# Record current action
self.CHO_hist.append(CHO)
self.insulin_hist.append(insulin)
# Record next observation
self.time_hist.append(self.time)
self.BG_hist.append(BG)
self.CGM_hist.append(CGM)
self.risk_hist.append(risk)
self.LBGI_hist.append(LBGI)
self.HBGI_hist.append(HBGI)
# Compute reward, and decide whether game is over
window_size = int(60 / self.sample_time)
BG_last_hour = self.CGM_hist[-window_size:]
reward = reward_fun(BG_last_hour)
# Stopping criteria
# done = BG < 70 or BG > 350 # If the BG level go out of control
done = self.patient.t >= 1440 # At the end of the day (24hrs = 1440 mins)
obs = Observation(CGM=CGM)
return Step(
observation=obs,
reward=reward,
done=done,
sample_time=self.sample_time,
patient_name=self.patient.name,
meal=CHO,
patient_state=self.patient.state)
def _reset(self):
self.sample_time = self.sensor.sample_time
self.viewer = None
BG = self.patient.observation.Gsub
horizon = 1
LBGI, HBGI, risk = risk_index([BG], horizon)
CGM = self.sensor.measure(self.patient)
self.time_hist = [self.scenario.start_time]
self.BG_hist = [BG]
self.CGM_hist = [CGM]
self.risk_hist = [risk]
self.LBGI_hist = [LBGI]
self.HBGI_hist = [HBGI]
self.CHO_hist = []
self.insulin_hist = []
def neg_risk(BG_last_hour):
if len(BG_last_hour) < 2:
return 0
else:
_, _, risk_current = risk_index([BG_last_hour[-1]], 1)
return - risk_current