search

iamtrask / Grokking-Deep-Learning

this repository accompanies the book "Grokking Deep Learning"
7.42k stars 1.57k forks source link

Chapter 6: layer_2_delta calculation #37

Open Vaticinator opened 4 years ago

Vaticinator commented 4 years ago

Section "Backpropagation in Code": There is: layer_2_delta = (walk_vs_stop[i:i+1] - layer_2) I belive it should be: layer_2_delta = (layer_2 - walk_vs_stop[i:i+1])

RomanGhost commented 4 years ago

So it is possible, but for weight training you need + = instead of - =.

LittleTownStudio commented 4 years ago 
  layer_2_delta = (walk_vs_stop[i:i+1] - layer_2)
  layer_1_delta = layer_2_delta.dot(weights_1_2.T)*relu2deriv(layer_1)

  weights_1_2 += alpha * layer_1.T.dot(layer_2_delta)
  weights_0_1 += alpha * layer_0.T.dot(layer_1_delta)     

  #### is same #######

  layer_2_delta = (layer_2 - walk_vs_stop[i:i+1])
  layer_1_delta=layer_2_delta.dot(weights_1_2.T)*relu2deriv(layer_1)

  weights_1_2 -= alpha * layer_1.T.dot(layer_2_delta)
  weights_0_1 -= alpha * layer_0.T.dot(layer_1_delta)

 ###
the two code result is same, but  i think the last code is logic right.