Hello friends. In order to test my Octave algorithms I started to create my own datasets so I have invented this one in Excel (the function is shown in cell H3). There are 6 features but x4 and x5 are useless, this has been done intentionally to simulate real life examples where more features than required are selected. In the future I might try also with codependent features like x7=0.5*sqrt(x3)*sin(x2) or something. I have run the algo with 1 hidden layer and got the following graph of the Cost Function: The blue line is the Cost Function with 6 neurons in the hidden layer, the red line is the Cost Function behavior with 60 neurons in the hidden layer. What confuses me is the behavior of the red line after 23 iterations, the Cost Function raises again. Is this behavior normal? and what could be an explanation for that?. Besides that, Ive just signed up in Kaggle competitions but as far as Ive seen so far, they are for pros. For now Im trying the Titanic: Machine Learning from Disaster if anyone knows a better way to get started I would appreciate your feedback. Dont forget the main question regarding the Cost Function behavior. P.S. It is a least squares Cost Function.
Posted on: Thu, 04 Sep 2014 06:20:47 +0000
Trending Topics
Recently Viewed Topics
© 2015