mindspore/tests/st/ops/ascend/test_adam_weight_decay.py

114 lines
5.2 KiB
Python

# Copyright 2022 Huawei Technologies Co., Ltd
#
# 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
#
# http://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.
# ============================================================================
import copy
import numpy as np
import mindspore.nn as nn
from mindspore import Tensor, Parameter, context
from mindspore.ops import operations as P
from mindspore.nn.optim.adam import _update_run_op
class OriNet(nn.Cell):
"""Origin net uses _update_run_op"""
def __init__(self, decay_flag):
super(OriNet, self).__init__()
self.decay_flag = decay_flag
self.optim_filter = True
def construct(self, param, m, v, lr, beta1, beta2, eps, weight_decay, gradient):
next_param = _update_run_op(beta1, beta2, eps, lr, weight_decay, param, m, v, gradient,
self.decay_flag, self.optim_filter)
return next_param
class FissionNet(nn.Cell):
"""Fission net uses P.AdamWeightDecay()"""
def __init__(self):
super(FissionNet, self).__init__()
self.optim_filter = True
def construct(self, param, m, v, lr, beta1, beta2, eps, weight_decay, gradient):
if self.optim_filter:
adam = P.AdamWeightDecay()
next_param = adam(param, m, v, lr, beta1, beta2, eps, weight_decay, gradient)
return next_param
return gradient
def test_adam_weight_decay_fission_1_decay_flag_is_true():
"""
Feature: AdamWeightDecay op
Description: test the rightness of AdamWeightDecay kernel, decay_flag is true
Expectation: the output is wrong
"""
decay_flag = True # equivalent to weight_decay is not zero
weight_decay = Parameter(Tensor(np.array([0.9]).astype(np.float32)), name="weight_decay")
beta1 = Parameter(Tensor(np.array([0.9]).astype(np.float32)), name="beta1")
beta2 = Parameter(Tensor(np.array([0.999]).astype(np.float32)), name="beta2")
eps = Parameter(Tensor(np.array([1e-8]).astype(np.float32)), name="eps")
lr = Parameter(Tensor(np.array([0.001]).astype(np.float32)), name="lr")
gradient = Parameter(Tensor(np.array([[2, 3], [1, 5]]).astype(np.float32)), name="gradient")
# The inputs: param, m and v will be modified in-place by P.AdamWeightDecay() or _update_run_op(),
# so here defines two copied of them: (param1, m1, v1) and (param2, m2, v2)
param1 = Parameter(Tensor(np.array([[1, 2], [3, 4]]).astype(np.float32)), name="param1")
m1 = Parameter(Tensor(np.array([[5, 6], [7, 8]]).astype(np.float32)), name="m1")
v1 = Parameter(Tensor(np.array([[3, 1], [7, 4]]).astype(np.float32)), name="v1")
param2 = copy.deepcopy(param1)
m2 = copy.deepcopy(m1)
v2 = copy.deepcopy(v1)
context.set_context(mode=context.PYNATIVE_MODE, device_target='Ascend')
origin_net = OriNet(decay_flag)
output1 = origin_net(param1, m1, v1, lr, beta1, beta2, eps, weight_decay, gradient)
fission_net = FissionNet()
output2 = fission_net(param2, m2, v2, lr, beta1, beta2, eps, weight_decay, gradient)
assert (output1.asnumpy() == output2[0].asnumpy()).all()
def test_adam_weight_decay_fission_2_decay_flag_is_false():
"""
Feature: AdamWeightDecay op
Description: test the rightness of ScaleGrad kernel, decay_flag is false
Expectation: the output is wrong
"""
decay_flag = False # equivalent to weight_decay is zero
weight_decay = Parameter(Tensor(np.array([0]).astype(np.float32)), name="weight_decay")
beta1 = Parameter(Tensor(np.array([0.9]).astype(np.float32)), name="beta1")
beta2 = Parameter(Tensor(np.array([0.999]).astype(np.float32)), name="beta2")
eps = Parameter(Tensor(np.array([1e-8]).astype(np.float32)), name="eps")
lr = Parameter(Tensor(np.array([0.001]).astype(np.float32)), name="lr")
gradient = Parameter(Tensor(np.array([[2, 3], [1, 5]]).astype(np.float32)), name="gradient")
# The inputs: param, m and v will be modified in-place by P.AdamWeightDecay() or _update_run_op(),
# so here defines two copied of them: (param1, m1, v1) and (param2, m2, v2)
param1 = Parameter(Tensor(np.array([[1, 2], [3, 4]]).astype(np.float32)), name="param1")
m1 = Parameter(Tensor(np.array([[5, 6], [7, 8]]).astype(np.float32)), name="m1")
v1 = Parameter(Tensor(np.array([[3, 1], [7, 4]]).astype(np.float32)), name="v1")
param2 = copy.deepcopy(param1)
m2 = copy.deepcopy(m1)
v2 = copy.deepcopy(v1)
context.set_context(mode=context.PYNATIVE_MODE, device_target='Ascend')
origin_net = OriNet(decay_flag)
output1 = origin_net(param1, m1, v1, lr, beta1, beta2, eps, weight_decay, gradient)
fission_net = FissionNet()
output2 = fission_net(param2, m2, v2, lr, beta1, beta2, eps, weight_decay, gradient)
assert (output1.asnumpy() == output2[0].asnumpy()).all()