def __init__(self, patient_number=0):
"""
Initializing the simulation environment.
"""
np.random.seed(1) ### Fixing seed
self.previous_action = 0
## Loading variable parameters
reward_flag, bg_init_flag = self._update_parameters()
# Initialize bolus history
self.bolusHistoryIndex = 0
self.bolusHistoryValue = []
self.bolusHistoryTime = []
self.insulinOnBoard = np.zeros(1)
# Initialize sensor model
self.CGMlambda = 15.96 # Johnson parameter of recalibrated and synchronized sensor error.
self.CGMepsilon = -5.471 # Johnson parameter of recalibrated and synchronized sensor error.
self.CGMdelta = 1.6898 # Johnson parameter of recalibrated and synchronized sensor error.
self.CGMgamma = -0.5444 # Johnson parameter of recalibrated and synchronized sensor error.
self.CGMerror = 0
self.sensor_noise = np.random.randn(1)
# self.CGMaux = []
self.sensorNoiseValue = 0.07 # Set a value
# =======================
# Anas patient parameters
# =======================
P, init_basal, carb_factor, _ = matlab_to_python(patient_number)
init_basal_optimal = init_basal
self.P = P
self.init_basal_optimal = init_basal_optimal
self.bolus = carb_factor
self.action_space = spaces.Box(0, 2*self.init_basal_optimal[0], (1,), dtype=np.float32)
if bg_init_flag == 'random':
self.init_basal = np.random.choice(np.linspace(init_basal_optimal-2, init_basal_optimal, 10))
elif bg_init_flag == 'fixed':
self.init_basal = init_basal_optimal
# Flag for manually resetting the init
self.reset_basal_manually = None
self._seed()
self.viewer = None
# ==========================================
# Setting up the Hovorka simulator
# ==========================================
# Initial values for parameters
initial_pars = (self.init_basal, 0, P)
# Initial value
X0 = fsolve(hovorka_model_tuple, np.zeros(11), args=initial_pars)
self.X0 = X0
# Simulation setup
self.integrator = ode(hovorka_model)
self.integrator.set_integrator('vode', method='bdf', order=5)
self.integrator.set_initial_value(X0, 0)
# Simulation time in minutes
self.simulation_time = 30
self.n_solver_steps = 1
self.stepsize = int(self.simulation_time/self.n_solver_steps)
self.observation_space = spaces.Box(0, 500, (int(self.stepsize + 4 + 2),), dtype=np.float32)
# State is BG, simulation_state is parameters of hovorka model
initial_bg = X0[-1] * 18
initial_insulin = np.ones(4) * self.init_basal_optimal
initial_iob = np.zeros(1)
self.state = np.concatenate([np.repeat(initial_bg, self.simulation_time), initial_insulin, initial_iob, np.zeros(1)])
self.simulation_state = X0
# Keeping track of entire blood glucose level for each episode
self.bg_history = []
self.insulin_history = initial_insulin
# ====================
# Meal setup
# ====================
eating_time = 1
meals, meal_indicator = meal_generator(eating_time=eating_time, premeal_bolus_time=0)
self.meals = meals
self.meal_indicator = meal_indicator
self.eating_time = eating_time
# Counter for number of iterations
self.num_iters = 0
# If blood glucose is less than zero, the simulator is out of bounds.
self.bg_threshold_low = 0
self.bg_threshold_high = 500
# TODO: This number is arbitrary
self.max_iter = 2160
# Reward flag
self.reward_flag = reward_flag
self.steps_beyond_done = None
def __init__(self, patient_number=None):
"""
Initializing the simulation environment.
"""
# Action space
self.previous_action = 0
# State space
self.observation_space = spaces.Box(0, 500, (34,), dtype=np.float32)
# self.observation_space = spaces.Box(0, 500, 1)
self.bolus = 6
## Loading variable parameters
# meal_times, meal_amounts, reward_flag, bg_init_flag = self._update_parameters()
reward_flag, bg_init_flag = self._update_parameters()
# Action space
# ====================================
# Normalized action space!!
# ====================================
self.action_space = spaces.Box(0, 50, (1,), dtype=np.float32)
# self.action_space = spaces.Box(-1, 1, (1,))
# self.seed()
self.viewer = None
# ==========================================
# Setting up the cambridge simulator
# ==========================================
# Selecting the virtual patient parameter
if patient_number==None:
patient_number = 0
self.patient_number = patient_number
P = pars[:, patient_number]
self.init_basal = init_basal_rates[patient_number]
# Patient parameters -- sub_1() means virtual patient #1
# P = subject(1)
# P = subject(2)
# P = subject(3)
# P = subject(4)
# P = subject(5)
# P = subject(6)
self.P = P
# Initial basal -- this rate dictates the initial BG value
if bg_init_flag == 'random':
self.init_basal = np.random.choice(np.linspace(init_basal_rates[self.patient_number]-2, init_basal_rates[self.patient_number], 10))
# Flag for manually resetting the init
self.reset_basal_manually = None
# Initial values for parameters
initial_pars = (self.init_basal, 0, P)
# Initial value
X0 = fsolve(cambridge_model_tuple, np.zeros(11), args=initial_pars)
self.X0 = X0
# Simulation setup
self.integrator = ode(cambridge_model)
self.integrator.set_integrator('vode', method='bdf', order=5)
self.integrator.set_initial_value(X0, 0)
# Simulation time in minutes
self.simulation_time = 30
# State is BG, simulation_state is parameters of hovorka model
initial_bg = X0[-1] * 18
initial_insulin = np.zeros(4)
self.state = np.concatenate([np.repeat(initial_bg, self.simulation_time), initial_insulin])
self.simulation_state = X0
# Keeping track of entire blood glucose level for each episode
self.bg_history = []
self.insulin_history = initial_insulin
# ====================
# Meal setup
# ====================
eating_time = 30
meals, meal_indicator = meal_generator(eating_time=eating_time)
# TODO: Clean up these
self.meals = meals
self.meal_indicator = meal_indicator
self.eating_time = eating_time
# Counter for number of iterations
self.num_iters = 0
# If blood glucose is less than zero, the simulator is out of bounds.
self.bg_threshold_low = 0
self.bg_threshold_high = 500
# TODO: This number is arbitrary
self.max_iter = 1440
# Reward flag
self.reward_flag = reward_flag
self.steps_beyond_done = None
def __init__(self):
"""
Initializing the simulation environment.
"""
# Fixing the random seed -- for reproducible experiments
np.random.seed(1)
# Miguel: Why is this needed?
self.previous_action = 0
# Bolus carb factor -- [g/U]
# self.bolus = 30
self.bolus = 25
## Loading variable parameters -- used if environment is extended
reward_flag, bg_init_flag = self._update_parameters()
# Miguel: CamelCase is not used in rest of code
# Initialize bolus history -- used for insulin on board
self.bolusHistoryIndex = 0
self.bolusHistoryValue = []
self.bolusHistoryTime = []
self.insulinOnBoard = np.zeros(1)
# Initialize sensor model -- Miguel: do we need all of this?
# self.CGMlambda = 15.96 # Johnson parameter of recalibrated and synchronized sensor error.
# self.CGMepsilon = -5.471 # Johnson parameter of recalibrated and synchronized sensor error.
# self.CGMdelta = 1.6898 # Johnson parameter of recalibrated and synchronized sensor error.
# self.CGMgamma = -0.5444 # Johnson parameter of recalibrated and synchronized sensor error.
# self.CGMerror = 0
self.sensor_noise = np.random.randn(1)
# self.CGMaux = []
# self.sensorNoiseValue = 0.07 # Set a value
# Model parameters
P = hovorka_parameters(70)
init_basal_optimal = 6.43
self.P = P
self.init_basal_optimal = init_basal_optimal
# Initializing the action space
self.action_space = spaces.Box(0, 2*self.init_basal_optimal, (1,), dtype=np.float32)
# Initial basal rate -- used for init and reset. Either randomly initialized or by a fixed value.
if bg_init_flag == 'random':
self.init_basal = np.random.choice(np.linspace(init_basal_optimal-2, init_basal_optimal, 10))
elif bg_init_flag == 'fixed':
self.init_basal = init_basal_optimal
# Flag for manually resetting the init when the episode restarts
self.reset_basal_manually = None
self._seed()
self.viewer = None
# ==========================================
# Setting up the Hovorka simulator
# ==========================================
# Initial values for parameters
initial_pars = (self.init_basal, 0, P)
# Initial value
X0 = fsolve(hovorka_model_tuple, np.zeros(11), args=initial_pars)
self.X0 = X0
# Simulation setup
self.integrator = ode(hovorka_model)
self.integrator.set_integrator('vode', method='bdf', order=5)
self.integrator.set_initial_value(X0, 0)
# Simulation time in minutes -- the default is solving the simulator 30 minutes at a time
# with a solver time of one minute.
self.simulation_time = 30
self.n_solver_steps = 1
self.stepsize = int(self.simulation_time/self.n_solver_steps)
# State space for the environment -- [30 min of BG, last 4 insulin action, bolus (if given) during last 30 mins]
self.observation_space = spaces.Box(0, 500, (int(self.stepsize + 4 + 2),), dtype=np.float32)
# State is BG, simulation_state is parameters of hovorka model
initial_bg = X0[-1] * 18
initial_insulin = np.ones(4) * self.init_basal_optimal
initial_iob = np.zeros(1)
# Initial state
self.state = np.concatenate([np.repeat(initial_bg, self.stepsize), initial_insulin, initial_iob, np.zeros(1)])
self.simulation_state = X0
# Keeping track of entire blood glucose level and insulin for each episode
self.bg_history = []
self.insulin_history = initial_insulin
# ====================
# Meal setup
# ====================
# Default eating time is considered one minute -- empirically the simulations are not too sensitive to this choice.
eating_time = 1
# Meals are carb intake and meal_indicator is the counted carbs by the patient
meals, meal_indicator = meal_generator(eating_time=eating_time, premeal_bolus_time=0)
self.meals = meals
self.meal_indicator = meal_indicator
self.eating_time = eating_time
# Counter for number of iterations
self.num_iters = 0
# If blood glucose is less than zero or above 500, the simulator is considered out of bounds.
self.bg_threshold_low = 0
self.bg_threshold_high = 500
# The max episode lenght is 36 hours
self.max_iter = 2160
# Reward flag
self.reward_flag = reward_flag
self.steps_beyond_done = None