Blood Glucose Level Prediction using Stochastic Gradient Descent

# Blood Glucose Level Prediction using Stochastic Gradient Descent
# Yong-Jun Shin (2021)

import numpy as np
import matplotlib.pyplot as plt

N = 300                  # number of total glucose measurements
n = np.arange(0, N, 1)   # [0,..., N-1] (vector)
g = np.empty(N)          # measured glucose concentration in mg/dL (vector)

# import glucose measurement data
count = 0
with open ('CGM.txt', 'r') as file:
  next(file)
  for line in file:
    line_list = line.split('|')
    g[count] = line_list[3]
    count = count + 1

g_est = np.zeros(N)      # estimated glucose concentration in ug/dL (vector)  g_est --> glucose 추정값
P1_opt = np.zeros(N)     # estimated P1 parameter (vector)                                                          
P2_opt = np.zeros(N)     # estimated P2 parameter (vector)                                                        
e = np.zeros(N)          # glucose concentration estimation error in uM (vector) (= y - y_est)
P1_opt[2] = 0.1          # initial estimated P1 parameter value                                                   
P1_opt[2] = 0.2          # initial estimated P2 parameter value
s = 0.000001             # step size s

for i in range (2, N-2):                             
    # parameter estimation using stochastic gradient descent                           
    g_est[i] = P1_opt[i]*g[i-1] + P2_opt[i]*g[i-2]       # estimated glucose concentration
    e[i] = g[i] - g_est[i]                               # glucose concentration estimation error
    P1_opt[i+1] = P1_opt[i] + s*g[i-1]*e[i]              # estimated P1 parameter
    P2_opt[i+1] = P2_opt[i] + s*g[i-2]*e[i]              # estimated P2 parameter
    
plt.plot(n, g, 'g', label = 'measured glucose level')                                         
plt.plot(n, g_est, 'b', label = 'estimated glucose level')
plt.plot(n, e, 'r', label = 'glucose level estimation error')
plt.xlabel('time (i)')                                                         
plt.ylabel('blood glucose concentration (mg/dL)')
plt.legend(loc='upper left')
plt.title('Blood Glucose Level Prediction using Stochastic Gradient Descent ')
plt.axis([0, N, -50, 400])
plt.grid(True)
plt.show()