overfitting

[JasLin]

@cloggy45 i just read the source of neuralnet version strategy.

let layers = [
      {type:'input', out_sx:1, out_sy:1, out_depth: 1},
      {type:'fc', num_neurons: this.layer_neurons, activation: this.layer_activation},
      {type:'regression', num_neurons: 1}
    ];

in theory, if we only use fully connected layer without doing any thing like regularization or dropout , the trained network is easily overfitted

[JasLin]

convnetjs haven't implement dropout yet. ref: karpathy/convnetjs#3

[GO1984]

are you shure? I found this snippet:
// dropout half the units (probability 0.5) in this layer during training, for regularization {type:'fc', num_neurons:10, activation:'relu', drop_prob: 0.5}
https://cs.stanford.edu/people/karpathy/convnetjs/docs.html

[JasLin]

@GO1984, hi bro, great, u r right, the ref I had gave is about DropConnect, but not dropout, they are difference things.

maybe we should add this into the layer design, and backtest it.

[cloggy45]

Any changes you would like to make to strategy submit a PR. In the meantime I will close this issue.