Optimizer

Optimizer

class tflibs.training.optimizer.Optimizer(learning_rate, var_scope, beta1, beta2, decay_policy='none', decay_params=None)

apply_gradients(grads_and_vars, global_step=None)

apply_tower_gradients(tower_grads, global_step=None)

Returns average gradients.

Parameters:tower_grads (list) – A list of tuples containing gradients and corresponding variables. Returns:Applying gradients operation Return type:tf.Operation

compute_grad(loss)

Computes gradients of trainable variables with regard to a loss given.

Parameters:loss (tf.Tensor) – A tf.Tensor of a loss. Returns:A list of tuples containing gradients and corresponding variables. Return type:list

static dying_decay(train_iters, decay_iters, decay_steps)

static lambda_decay(lr_fn)

static step_decay(train_iters, decay_steps, decay_rate)

train_op(loss, global_step=None)

var_list