import jpype
from neuralogic.optim.lr_scheduler.lr_decay import LRDecay
[docs]
class GeometricLR(LRDecay):
r"""
Decay learning rate on every :math:`steps` epoch by the following formula
.. math::
\mathbf{lr}_i = \mathbf{lr}_{i-1} \cdot decay\_rate
Parameters
----------
decay_rate : float
steps : int
"""
def __init__(self, decay_rate: float, steps: int):
super().__init__()
self.decay_rate = decay_rate
self.steps = steps
def _initialize(self, learning_rate):
class_name = "cz.cvut.fel.ida.neural.networks.computation.training.strategies.Hyperparameters.GeometricDecay"
self._decay = jpype.JClass(class_name)(learning_rate, self.decay_rate, self.steps)
return self._decay
def __str__(self) -> str:
return f"GeometricLR(decay_rate={self.decay_rate}, steps={self.steps})"