optim_adagrad: Adagrad optimizer

In torch: Tensors and Neural Networks with 'GPU' Acceleration

Adagrad optimizer

Description

Proposed in Adaptive Subgradient Methods for Online Learning and Stochastic Optimization

Usage

optim_adagrad(
  params,
  lr = 0.01,
  lr_decay = 0,
  weight_decay = 0,
  initial_accumulator_value = 0,
  eps = 1e-10
)

Arguments

params	(iterable): list of parameters to optimize or list parameter groups
lr	(float, optional): learning rate (default: 1e-2)
lr_decay	(float, optional): learning rate decay (default: 0)
weight_decay	(float, optional): weight decay (L2 penalty) (default: 0)
initial_accumulator_value	the initial value for the accumulator. (default: 0) Adagrad is an especially good optimizer for sparse data. It individually modifies learning rate for every single parameter, dividing the original learning rate value by sum of the squares of the gradients. It causes that the rarely occurring features get greater learning rates. The main downside of this method is the fact that learning rate may be getting small too fast, so that at some point a model cannot learn anymore.
eps	(float, optional): term added to the denominator to improve numerical stability (default: 1e-10)

Warning

If you need to move a model to GPU via ⁠$cuda()⁠, please do so before constructing optimizers for it. Parameters of a model after ⁠$cuda()⁠ will be different objects from those before the call. In general, you should make sure that the objects pointed to by model parameters subject to optimization remain the same over the whole lifecycle of optimizer creation and usage.

Note

Update rule:

\theta_{t+1} = \theta_{t} - \frac{\eta }{\sqrt{G_{t} + \epsilon}} \odot g_{t}

The equation above and some remarks quoted after An overview of gradient descent optimization algorithms by Sebastian Ruder.

torch documentation built on June 7, 2023, 6:19 p.m.