Wrong calculation in gradient descent about deep-learning HOT 2 CLOSED

I think it's the same formula as long as d_error = -(y - output), so

weights = weights - learnrate * d_error * x = 
= weights - learnrate * (-(y - output)) * x =
= weights + learnrate * (y - output) * x

The same for the bias term.

from deep-learning.

Closing this issue as the mathematical equation and the implementation is exactly the same.

If d_error = -(y - output)

weights -= learnrate * d_error * x
bias -= learnrate * d_error

If d_error = (y - output)

weights += learnrate * d_error * x
bias += learnrate * d_error

from deep-learning.