Open
Description
Why you multiply the sigmoid_prime(x) by x? You defined backpropagation error as (π¦βπ¦Μ)πβ²(π₯)
in the notebook:
Your solution:
def error_term_formula(x, y, output):
# for binary cross entropy loss
return (y - output)*x
# for mean square error
# return (y - output)*sigmoid_prime(x)*x
Notebook definition:
TODO: Backpropagate the error
Now it's your turn to shine. Write the error term. Remember that this is given by the equation
(π¦βπ¦Μ)πβ²(π₯)
You even had
(y - output)*sigmoid_prime(x)
as the solution, but then you changed it to(y - output)*sigmoid_prime(x)*x
in this commit
Metadata
Metadata
Assignees
Labels
No labels