• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 2021 Huawei Technologies Co., Ltd
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7# http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14# ============================================================================
15import os
16import numpy as np
17import pytest
18
19import mindspore.context as context
20import mindspore.nn as nn
21from mindspore import Tensor
22from mindspore.nn import TrainOneStepCell, WithLossCell
23from mindspore.nn.optim import Momentum
24from mindspore.ops import operations as P
25
26context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", save_compile_cache=True, load_compile_cache=True)
27
28
29class LeNet(nn.Cell):
30    def __init__(self):
31        super(LeNet, self).__init__()
32        self.relu = P.ReLU()
33        self.batch_size = 32
34
35        self.conv1 = nn.Conv2d(1, 6, kernel_size=5, stride=1, padding=0, has_bias=False, pad_mode='valid')
36        self.conv2 = nn.Conv2d(6, 16, kernel_size=5, stride=1, padding=0, has_bias=False, pad_mode='valid')
37        self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
38        self.reshape = P.Reshape()
39        self.fc1 = nn.Dense(400, 120)
40        self.fc2 = nn.Dense(120, 84)
41        self.fc3 = nn.Dense(84, 10)
42
43    def construct(self, input_x):
44        output = self.conv1(input_x)
45        output = self.relu(output)
46        output = self.pool(output)
47        output = self.conv2(output)
48        output = self.relu(output)
49        output = self.pool(output)
50        output = self.reshape(output, (self.batch_size, -1))
51        output = self.fc1(output)
52        output = self.relu(output)
53        output = self.fc2(output)
54        output = self.relu(output)
55        output = self.fc3(output)
56        return output
57
58
59def train(net, data, label):
60    learning_rate = 0.01
61    momentum = 0.9
62
63    optimizer = Momentum(filter(lambda x: x.requires_grad, net.get_parameters()), learning_rate, momentum)
64    criterion = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
65    net_with_criterion = WithLossCell(net, criterion)
66    train_network = TrainOneStepCell(net_with_criterion, optimizer)  # optimizer
67    train_network.set_train()
68    res = train_network(data, label)
69    print("+++++++++Loss+++++++++++++")
70    print(res)
71    print("+++++++++++++++++++++++++++")
72    diff = res.asnumpy() - 2.302585
73    assert np.all(diff < 1.e-6)
74
75
76@pytest.mark.level0
77@pytest.mark.platform_x86_ascend_training
78@pytest.mark.platform_arm_ascend_training
79@pytest.mark.env_onecard
80def test_lenet():
81    path = "compile_cache.mindir"
82    if os.path.exists(path):
83        os.remove(path)
84    assert not os.path.exists(path)
85    data = Tensor(np.ones([32, 1, 32, 32]).astype(np.float32) * 0.01)
86    label = Tensor(np.ones([32]).astype(np.int32))
87    net = LeNet()
88    train(net, data, label)
89    assert os.path.exists(path)
90
91    data1 = Tensor(np.ones([32, 1, 32, 32]).astype(np.float32) * 0.01)
92    label1 = Tensor(np.ones([32]).astype(np.int32))
93    net1 = LeNet()
94    train(net1, data1, label1)
95    context.set_context(save_compile_cache=False, load_compile_cache=False)
96