Skip to content

Home

Jump to bottom
Inyong Hwang edited this page Jun 14, 2019 · 1 revision

Welcome to the Study wiki!

Optimizers

  • Parameters

    • clipnorm=x: all parameter gradients clip to a maximum norm of x.
    • clipvalue=x: all parameter gradients clip to a max value of x and min value of -x.

  • SGD (Stochastic Gradient Descent)

keras.optimizers.SGD(lr, momentum, decay, nesterov)

- Arguments
    - lr: Learning rate, float >= 0.
    - momentum: Parameter that accelerates SGD in the relevant direction and dampens oscillations, float >= 0.
    - decay: Learning rate decay over each update, float >= 0.
    - nesterov: Whether to apply Nesterov momentum, boolean.
  • RMSprop
keras.optimizers.RMSprop(lr, rho, epsilon, decay)
  • Adagrad
keras.optimizers.Adagrad(lr, epsilon, decay)
  • Adadelta
keras.optimizers.Adadelta(lr, rho, epsilon, decay=)
  • Adam
keras.optimizers.Adam(lr, beta_1, beta_2, epsilon, decay, amsgrad)
  • Adammax
keras.optimizers.Adamax(lr, beta_1, beta_2, epsilon, decay)
  • Nadam
keras.optimizers.Nadam(lr, beta_1, beta_2, epsilon, schedule_decay)

KERAS

Clone this wiki locally