forked from mindspore-Ecosystem/mindspore
130 lines
5.3 KiB
Python
130 lines
5.3 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 numpy as np
|
|
import mindspore.context as context
|
|
from mindspore import nn, Tensor
|
|
from optimizer_utils import build_network, \
|
|
loss_default_adamax, loss_not_default_adamax, loss_group_adamax
|
|
|
|
|
|
w1 = np.array([[0.03909272, 0.08893055, -0.259909, -0.459185,
|
|
-0.0195536, 0.12977135, -0.62942827, -0.53132117],
|
|
[0.1542052, 0.6513571, -0.06453168, 0.44788414,
|
|
-0.3775454, 0.6520292, 0.444174, -0.59306043],
|
|
[0.2712369, 0.20890862, 0.6859066, 0.6629662,
|
|
0.4724893, -0.34384444, -0.16007674, 0.21797538],
|
|
[-0.3865972, 0.26727962, 0.23178828, -0.24629539,
|
|
-0.68038213, -0.31262863, 0.10493469, -0.28973007]]).astype("float32")
|
|
|
|
b1 = np.array([0., 0., 0., 0.]).astype("float32")
|
|
|
|
w2 = np.array([[-0.6079024, -1.005364, 0.59004724, 0.7289244]]).astype("float32")
|
|
|
|
b2 = np.array([0.]).astype("float32")
|
|
|
|
|
|
class Net(nn.Cell):
|
|
"""
|
|
build a 2-layers net to test adamax optimizer
|
|
"""
|
|
def __init__(self):
|
|
super(Net, self).__init__()
|
|
self.fc1 = nn.Dense(8, 4, weight_init=Tensor(w1), bias_init=Tensor(b1))
|
|
self.fc2 = nn.Dense(4, 1, weight_init=Tensor(w2), bias_init=Tensor(b2))
|
|
self.relu = nn.ReLU()
|
|
|
|
def construct(self, x):
|
|
x = self.relu(self.fc1(x))
|
|
return self.fc2(x)
|
|
|
|
|
|
def test_default_adamax_pynative():
|
|
"""
|
|
Feature: Test adamax optimizer
|
|
Description: Test adamax in Pynative mode with default parameter
|
|
Expectation: Loss values and parameters conform to preset values.
|
|
"""
|
|
context.set_context(mode=context.PYNATIVE_MODE, device_target='Ascend')
|
|
config = {'name': 'adamax', 'lr': 0.001, "beta1": 0.9, "beta2": 0.999, "eps": 1e-07,
|
|
'weight_decay': 0.0}
|
|
loss = build_network(config, net=Net(), loss_fn=nn.MSELoss(reduction='mean'))
|
|
assert np.allclose(loss_default_adamax, loss, atol=1.e-5)
|
|
|
|
|
|
def test_default_adamax_graph():
|
|
"""
|
|
Feature: Test adamax optimizer
|
|
Description: Test adamax in Graph mode with default parameter
|
|
Expectation: Loss values and parameters conform to preset values.
|
|
"""
|
|
context.set_context(mode=context.GRAPH_MODE, device_target='Ascend')
|
|
config = {'name': 'adamax', 'lr': 0.001, "beta1": 0.9, "beta2": 0.999, "eps": 1e-07,
|
|
'weight_decay': 0.0}
|
|
loss = build_network(config, net=Net(), loss_fn=nn.MSELoss(reduction='mean'))
|
|
assert np.allclose(loss_default_adamax, loss, atol=1.e-5)
|
|
|
|
|
|
def test_no_default_adamax_pynative():
|
|
"""
|
|
Feature: Test adamax optimizer
|
|
Description: Test adamax in Pynative mode with another set of parameter
|
|
Expectation: Loss values and parameters conform to preset values.
|
|
"""
|
|
context.set_context(mode=context.PYNATIVE_MODE, device_target='Ascend')
|
|
config = {'name': 'adamax', 'lr': 0.01, "beta1": 0.9, "beta2": 0.98, "eps": 1e-06,
|
|
'weight_decay': 0.0}
|
|
loss = build_network(config, net=Net(), loss_fn=nn.MSELoss(reduction='mean'))
|
|
assert np.allclose(loss_not_default_adamax, loss, atol=1.e-5)
|
|
|
|
|
|
def test_no_default_adamax_graph():
|
|
"""
|
|
Feature: Test adamax optimizer
|
|
Description: Test adamax in Graph mode with another set of parameter
|
|
Expectation: Loss values and parameters conform to preset values.
|
|
"""
|
|
context.set_context(mode=context.GRAPH_MODE, device_target='Ascend')
|
|
config = {'name': 'adamax', 'lr': 0.01, "beta1": 0.9, "beta2": 0.98, "eps": 1e-06,
|
|
'weight_decay': 0.0}
|
|
loss = build_network(config, net=Net(), loss_fn=nn.MSELoss(reduction='mean'))
|
|
assert np.allclose(loss_not_default_adamax, loss, atol=1.e-5)
|
|
|
|
|
|
def test_default_adamax_group_pynative():
|
|
"""
|
|
Feature: Test adamax optimizer
|
|
Description: Test adamax in Pynative mode with parameter grouping
|
|
Expectation: Loss values and parameters conform to preset values.
|
|
"""
|
|
context.set_context(mode=context.PYNATIVE_MODE, device_target='Ascend')
|
|
config = {'name': 'adamax', 'lr': 0.002, "beta1": 0.9, "beta2": 0.999, "eps": 1e-08,
|
|
'weight_decay': 0.0}
|
|
loss = build_network(config, is_group=True, net=Net(), loss_fn=nn.MSELoss(reduction='mean'))
|
|
assert np.allclose(loss_group_adamax, loss, atol=1.e-5)
|
|
|
|
|
|
def test_default_adamax_group_graph():
|
|
"""
|
|
Feature: Test adamax optimizer
|
|
Description: Test adamax in Graph mode with parameter grouping
|
|
Expectation: Loss values and parameters conform to preset values.
|
|
"""
|
|
context.set_context(mode=context.GRAPH_MODE, device_target='Ascend')
|
|
config = {'name': 'adamax', 'lr': 0.002, "beta1": 0.9, "beta2": 0.999, "eps": 1e-08,
|
|
'weight_decay': 0.0}
|
|
loss = build_network(config, is_group=True, net=Net(), loss_fn=nn.MSELoss(reduction='mean'))
|
|
assert np.allclose(loss_group_adamax, loss, atol=1.e-5)
|