vkML 0.1.0

RMSProp

RMSProp, matching torch.optim.RMSprop.

class RMSProp : Optimizerpython/vkml/optim.py:202

Divides the gradient by a running root-mean-square of recent gradients.

The running average starts at zero, as torch's does, so the first step is (1−α)·g² rather than . That difference persists for many steps through the exponential average, so it is not a detail — a reimplementation that initialises from the first gradient diverges from torch for a long time.

Construction

__init__

def __init__self, params, lr: float=0.01, alpha: float=0.99, eps: float=1e-08, weight_decay: float=0.0, momentum: float=0.0, centered: bool=Falsepython/vkml/optim.py:210

Update

See also Adam, SGD, Optimizer

vkML — Vulkan-first machine learning in C++20. Apache-2.0. Signatures on this page are generated from the installed module.