What's wrong with the following gradient descent algorithm? | Smarter Dev | Page 1

full bluff Mar 16, 2023, 12:54 PM

#

So, this is a simple gradient descent function I made for the cost function in linear regression and there are a few problems with it --
It's taking way too long to converge (70,000 iterations) for small datasets and it gives me an overflow error if the dataset is large with multiple input variables.

spring violet Mar 16, 2023, 3:02 PM

#

What cost function are you using?

full bluff Mar 16, 2023, 5:56 PM

#

spring violet What cost function are you using?

Mean squared error

spring violet Mar 17, 2023, 5:05 AM

#

full bluff Mean squared error

Can you try using db = np.mean(loss)?

#

Oh wait your thing is different

#

Why is your bias a vector and not a scalar?

#

Oh wait it's the same

#

I think your learning rate is just too low

#What's wrong with the following gradient descent algorithm?