data_mining:neural_network:debugging

Gradient checking

Approximating derivatives:

(Large triangle (+/- triangle, two-sided difference))

$\frac{f(\Theta + \epsilon) - f(\Theta - \epsilon)}{2 \epsilon} \approx g(\Theta)$

$f'(\Theta) = \lim_{\epsilon->0} \frac{f(\Theta + \epsilon) - f(\Theta - \epsilon)}{2 \epsilon}$

Approx error is in $O(\epsilon^2)$

Take $W^{[1]}, b^{[1]}, \dots, W^{[L]},b^{[L]}$ and put it in a big vector $\theta$.

Take $dW^{[1]}, db^{[1]}, \dots, dW^{[L]},db^{[L]}$ and put it in a big vector $d\theta$.

J is now $J(\Theta) = J(\Theta_1, ...)$

For each i:

$d\Theta_{approx}[i] = \frac{J(\dots, \Theta_i+\epsilon,\dots) - J(\dots, \Theta_i-\epsilon,\dots)}{2\epsilon}$

$\epsilon = 10^{-7}$

  • data_mining/neural_network/debugging.txt
  • Last modified: 2017/08/20 00:12
  • by phreazer