C1_W1_Lab04_Gradient_Descent: Overflow Error

Course Q&A Machine Learning Specialization Supervised ML: Regression and Classification
1

Hello everyone, I was trying to compute the gradient descent on a different dataset and run into the error below.

C:\Users\migas\AppData\Local\Temp\ipykernel_34976\4288327375.py:11: RuntimeWarning: overflow encountered in double_scalars w_adj = (y_pred - y[i]) * x[i] C:\Users\migas\AppData\Local\Temp\ipykernel_34976\3680357176.py:12: RuntimeWarning: invalid value encountered in double_scalars w = w - learning_rate * dw C:\Users\migas\AppData\Local\Temp\ipykernel_34976\3680357176.py:13: RuntimeWarning: invalid value encountered in double_scalars b = b - learning_rate * db

I believe this overflow error is caused by the value being out of the defined range of the data type. I tried using numpy.exp() but it just rounded my results to 0. Is there something I can do here?

See the code below!

#### Computing the gradient

def compute_gradient(x,y,w,b):
    
(code removed)

w, b = 0,0

compute_gradient(x_org,y_org,w,b)

Output for one iteration → (-5.228267343780145e+149, -1.0251504595647343e+148)
`

#### Compute the gradient descent

def compute_gradient_descent(x,y,w_in,b_in,learning_rate ,num_iter):
(code removed)

w_final, b_final = compute_gradient_descent(x_org,y_org,w,b,0.01,10000)

2

Your compute_gradient() code should return the gradients - not compute new w and b values.

Updating the w and b values is the job of the compute_gradient_descent() function.

3

Also, normalize your features if any of them spans a wide range.

4

Thank you very much for your help, I can see what’s happening!

5

I had an outlier in my data that I have dealt with. Thank you!

1 Like