Rolodex
September 7, 2019, 7:19pm
1
Any idea about why, on the neuralnet
package, the sse
function for parameter: err.fct
divides by 2 the Sum of Squared Errors
?
From the document:
we have the following text:
we can ignore the Sum
of H
(outputs) and assume we only have just one output (H = 1). L corresponds to the observations.
In multiple bibliographies on internet they don't divide by 2.
Just in case, here is the neuralnet
package reference:
<p>Train neural networks using backpropagation,
resilient backpropagation (RPROP) with (Riedmiller, 1994) or without weight
backtracking (Riedmiller and Braun, 1993) or the modified globally
convergent version (GRPROP) by Anastasiadis et al. (2005)....
Thanks!
I cannot say whether there are some specific reasons for neuralnet
package, but usually this is done convenience.
When you'll be taking the derivative, you'll get a 2
from the square term, and if you used the function with 2
in the denominator, it gets adjusted with it.
Here's a related thread:
1 Like
Rolodex
September 8, 2019, 8:59pm
3
thank you @Yarnabrina , that helped!
system
Closed
September 15, 2019, 8:59pm
4
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.