# Copyright 2020 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
__all__ = ['SGD']
from typing import List
from objax.module import Module, ModuleList
from objax.typing import JaxArray
from objax.util import class_name
from objax.variable import TrainRef, TrainVar, VarCollection
[docs]
class SGD(Module):
"""Stochastic Gradient Descent (SGD) optimizer."""
[docs]
def __init__(self, vc: VarCollection):
"""Constructor for SGD optimizer.
Args:
vc: collection of variables to optimize.
"""
self.train_vars = ModuleList(TrainRef(x) for x in vc.subset(TrainVar))
[docs]
def __call__(self, lr: float, grads: List[JaxArray]):
"""Updates variables based on SGD algorithm.
Args:
lr: the learning rate.
grads: the gradients to apply.
"""
assert len(grads) == len(self.train_vars), 'Expecting as many gradients as trainable variables'
for g, p in zip(grads, self.train_vars):
p.value -= lr * g
def __repr__(self):
return f'{class_name(self)}()'