import numpy as np
alpha = 0.1
h=0.000001
ftol = 1e-12
def f(p):
_sum = 0 # simple sum of x and y squared
for i in range(len(p)):
_sum += p[i]**2
return _sum
def del_f(p):
dp = np.zeros_like(p)
print(f'p={p}')
for i in range(len(p)):
_save = p[i] # save p[i] so it can be restored later
p[i] += h # take a step in the positive direction
print(f'p={p}')
fhi = f(p) # evaluate after taking a positive step
print(f'fhi={fhi}')
p[i] = _save # restore p[i]
print(f'p={p}') # p[0] =
p[i] -= h # take a step in the negative direction.
print(f'p={p}') # p[0] shouldn't be zero
flo = f(p) # evaluate after taking a negative step
p[i] = _save # restore p[i]
print(f'flo={flo}')
p[i] = _save # restore cell to its original value
dp[i] = (fhi-flo)/(2*h) # calculate the gradient with P as the center value
print(f'dp={dp}')
pass
return dp
p = np.array([1,1])
while f(p) > ftol:
dp = del_f(p) # the gradient
p = p - alpha*dp # Now update the parameter values
print(f'p[0]={p[0]:11.9f} p[1]={p[1]:11.9f} dp[0]={dp[0]:8.6f} dp[1]={dp[1]:8.6f}')This is the start if a gradient descent program. There is a line above p[0] -= h tha sets p[0] to 0 when p[0]=1 in the line before.
After subtracting h p[0] should be equal to 0.99999 but instead is 0
This is the print out I get
p=[1 1]
p=[1 1]
fhi=2
p=[1 1]
p=[0 1]
flo=1
dp=[500000 0]
Since initial p = [1,1] and I am only adding or subtracting h which is small, p[0] should not be zero until after many iterations.
I tried several keywords to pretty print the python script but none worked. I see I just had to wait for the preview.
