microsoftml.sgd_optimizer: Stochastic gradient descent
Usage
microsoftml.sgd_optimizer(learning_rate: numbers.Real = None,
momentum: numbers.Real = None, nag: bool = None,
weight_decay: numbers.Real = None,
l_rate_red_ratio: numbers.Real = None,
l_rate_red_freq: numbers.Real = None,
l_rate_red_error_ratio: numbers.Real = None)
Description
Stochastic gradient descent optimizer.
Arguments
learning_rate
Learning rate (settings).
momentum
Momentum Term (settings).
nag
Use Nesterov's accelerated gradient (settings).
weight_decay
Weight decay (settings).
l_rate_red_ratio
Learning rate reduction ratio (settings).
l_rate_red_freq
Learning rate reduction ratio (settings).
l_rate_red_error_ratio
Relative error reduction criterion for learning rate reduction (settings).