LazyAdam

Customizes the LazyAdam optimizer.

def create_hash_optimizer(learning_rate=0.001, beta1=0.9, beta2=0.999, epsilon=1e-8, name="LazyAdam", use_fusion_optim=False)

Parameter	Type	Mandatory/Optional	Description
learning_rate	float/tf.Tensor	Optional	Learning rate Default value: 0.001 Value range: [0.0, 10.0]
beta1	float	Optional	Exponential decay rate estimated at the first moment Default value: 0.9 Value range: (0.0, 1.0)
beta2	float	Optional	Exponential decay rate estimated at the second moment Default value: 0.999 Value range: [0.0, 1.0]
epsilon	float	Optional	Adds the value to the denominator to improve data stability. Default value: 1e-8 Value range: (0.0, 1.0]
name	string	Optional	Name of the optimizer Default value: LazyAdam Name length range: [1, 200].
use_fusion_optim	bool	Optional	Whether to use the LazyAdam fused operator to calculate and update slot_m, slot_v, and variable. Default value: False Value range: True: The fused operator is used. You need to manually compile and deploy the LazyAdam fused operator. For details about how to use the operator and its restrictions, see README. False: The LazyAdam fused operator is not used.

An instance object of the CustomizedLazyAdam optimizer.

from mx_rec.optimizers.lazy_adam import create_hash_optimizer
hashtable_optimizer = create_hash_optimizer()

Parent topic: Optimizers