1

I am very new to Neural network and python and only know about very basic stuff. I tried to train the neural network with four categories of data however, when tested, the neural network tend to get one specific or two specific categories completely right and the other three categories completely wrong. I end up getting performance 0.25 /0.5. I am pretty sure my coding for the neural network is correct because it worked when I tried handwritten digits. I am just wondering under what kind of circumstance will this happen and if it is my datasets causing the problem.

user263008
  • 11
  • 1
  • Welcome to CV. Is there a class imbalance? In other words, is each category approximately equally represented? If not, you could try weighting your loss to offset the imbalance. – Frans Rodenburg Oct 18 '19 at 03:16
  • 1
    Thank you for the reply Frans. But I don't think it is class imbalance because my training dataset is evenly divided with these four categories. It is really weird that I run the code several times and every time the 'complete right' category is different, like for the first time it is A, then next time it will be C. – user263008 Oct 18 '19 at 03:27
  • 3
    Have a look at this question: [What should I do when my neural network doesn't learn?](https://stats.stackexchange.com/questions/352036/what-should-i-do-when-my-neural-network-doesnt-learn) – Frans Rodenburg Oct 18 '19 at 03:37

0 Answers0