Linear Activation Leads to NaN minmax
GoogleCodeExporter opened this issue · comments
Slightly modified standard feed-forward example
x = np.linspace(-7,7,20)
y = np.sin(x) * .5
size = len(x)
inp = x.reshape(size,1)
tar = y.reshape(size,1)
net = nl.net.newff([[-7,7]], [5,1], transf=[nl.net.trans.PureLin()]*2)
Leads to infinite minmax in core.py:
self.init() # line 97, minmax = [[-inf inf]]
Which leads to a problem in init.py, line 129/130
x = 2. / (minmax[:, 1] - minmax[:, 0])
y = 1. - minmax[:, 1] * x
Original issue reported on code.google.com by MLotst...@gmail.com
on 9 Jun 2014 at 10:00
Fix in trunk. I made replacement -inf/inf to -1/1, but may be Nguyen-Widrow is
not best method for init layers with linear actuator....
Thanks for your report
Original comment by zue...@gmail.com
on 10 Jun 2014 at 6:06
Original comment by zue...@gmail.com
on 10 Jun 2014 at 6:06
- Changed state: Fixed