jypyter notebook一个问题

import numpy as np
X = np.random.rand(200, 16)
w = np.random.randint(1, 100, size=(16, 1))
b = np.random.randint(1, 600, size=1)
y = X.dot(w) + b + np.random.randn(200, 1)
X = np.concatenate([X,np.full(shape=(200, 1), fill_value=1)], axis=1)
epoches = 10000
eta = 0.001
t0 = 5
t1 = 1000
def learning_rate_shedule(t):
    return t0 / (t + t1)
theta = np.random.randn(17,1)
t = 0
for i in range(epoches):
    g = X.T.dot(X.dot(theta)-y)
    eta = learning_rate_shedule(t)
    theta = theta - eta * g
    t +=1
print('真实的斜率、截距是:', w, b)
print('BGD求解的斜率、截距是:', theta)

执行结果是：

真实的斜率、截距是: [[10]
 [ 2]
 [40]
 [48]
 [57]
 [89]
 [ 6]
 [70]
 [67]
 [33]
 [70]
 [93]
 [84]
 [41]
 [61]
 [62]] [527]
BGD求解的斜率、截距是: [[nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]
 [nan]]

请问到底怎么修改呢，在15个参数情况下正常执行，到15以上就是NAN了