degitx / nen Goto Github PK

Hi Alexey, I'd like to ask you whether I need to scale down the input data between 0 and 1 before feeding it into the network or I can feed any size of numbers? Thanks.

Hi Alexey, could you kindly help me?

I'm training a network and it always gives an output of 1. No matter how I change the number of inputs (10, 100 or 500) or hidden neurons (2 or 20). There is 1 hidden layer and I'm using default values for learning rate etc.

I've got a training data consisting of groups with 100 numbers so I'd like to choose an input layer with 100 neurons. I train the network with positive set only, setting the output to 1 in every case. So when I finish training, I test the network with random input. I would expect that the network would predict a very low value since the random input differs entirely from my training data. But I always get 1.

Training it with random data always gives 1 either.

I also tested to train it with a single data to leave it untrained and when feeding random data in it, still get an output of 1. I would expect random output since the network is not trained yet and I'd believe that the hidden weights should still have random values.

What can be the problem? Thank you in advance.