20
Pizza
6y

My shitty neural network doesn't want to learn.

Wondering if the problem is in my implementation of the backpropagation, or in the fucking hyperparameters.

Stupid network go fuck yourself and suck my neuron.

Comments
  • 7
    This is what I expect to hear if more parents were honest about their children.
  • 2
    Probably hyperparamaters. It took me a month of grid and random searches to train mines. :( Gl !!
  • 0
    @penderis fuck yea, gona use that some day.
  • 1
    Maybe post the architecture and we can try to figure it out.Or maybe it's on strike :)
  • 0
    @bdhobare
    It's trying to learn the MONK's-1 dataset: https://archive.ics.uci.edu/ml/...
    (So a quite simple logic formula)

    1 hidden layer (tried 5 to 50 units), logistic activation functions
    Learning rate: 0.2 / 0.1 / 0.01
    Momentum: 0.1 / 0.01 / 0.001
    Regularization: 0.1 / 0.01 / 0.001
    Up to 1000 batch iterations

    Random weight initialization with a uniform distribution in [0,1] divided by fan-in.

    Training error is smooth but... Increasing :/
  • 1
    Your bug could be as small as a missing minus sign.... You should double check all of those
Add Comment