Machine Learning Skills

Aug 22, 2019


1. Feature Scaling


1566435926054

  • make gradient descent run much faster

  • converge in a lot fewer iterations

  • no need in normal equation method, but while using gradient descent, it’s still important

    1566436039525

    1566436108430

2. Learning Rate α


  • α is too small: slow convergence.
  • α is too large: may not decrease on every iteration and thus may not converge.

    1566437357098

3. Normal equation method


1566443319636

参考资料