forked from mindspore-courses/MindSpore-Tutorial
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
194 lines (148 loc) · 6.14 KB
/
main.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
"""MindSpore基本使用"""
import os
import tarfile
import urllib.request
import mindcv
import numpy as np
import mindspore
from mindspore import ops
from mindspore import nn
from mindspore.dataset import transforms
# ================================================================== #
# 内容表 #
# ================================================================== #
# 1. 自动微分例1 (Line 29 to 55)
# 2. 自动微分例2 (Line 61 to 100)
# 3. 从numpy中加载数据 (Line 106 to 113)
# 4. 输入 (Line 119 to 148)
# 5. 自定义数据集 (Line 154 to 158)
# 6. 预训练模型 (Line 164 to 177)
# 7. 保存和加载模型 (Line 183 to 189)
# ================================================================== #
# 1. 自动微分例 1 #
# ================================================================== #
# 创建Tensor
x_t = mindspore.Tensor(1.)
b_t = mindspore.Tensor(3.)
w_t = mindspore.Tensor(2.)
# 定义网络
class Net(nn.Cell):
"""f=wx+b"""
def __init__(self):
super().__init__()
self.b = mindspore.Tensor(3.)
self.w = mindspore.Tensor(2.)
def construct(self, x, w, b):
f = w * x + b
return f
# 计算梯度
grad_op = ops.GradOperation(get_all=True)
net = Net()
grad_fn = grad_op(net, weights=(x_t, w_t, b_t))
grads = grad_fn(x_t, w_t, b_t)
# 打印梯度
print(grads[0]) # x.grad = 2
print(grads[1]) # w.grad = 1
print(grads[2]) # b.grad = 1
# ================================================================== #
# 2. 自动微分例 2 #
# ================================================================== #
# 创建形状为(10,3)和(10,2)的随机Tensor.
x_t = ops.randn(10, 3)
y = ops.randn(10, 2)
# 构造全连接层
linear = nn.Dense(3, 2)
print('w: ', linear.weight)
print('b: ', linear.bias)
# 构造损失函数和优化器
criterion = nn.MSELoss()
optimizer = nn.optim.SGD(linear.trainable_params(), 0.01)
# 定义正向传播函数
def forward(x):
"""正向传播函数"""
pred = linear(x)
loss = criterion(pred, y)
return pred, loss
# 定义求梯度函数
grad_fn = ops.value_and_grad(forward, None, optimizer.parameters, has_aux=True)
# 正向传播
(pred_t, loss_t), grads = grad_fn(x_t)
# 打印loss.
print('loss: ', loss_t.asnumpy().item())
# 打印梯度
print('dL/dw: ', grads[0])
print('dL/db: ', grads[1])
optimizer(grads)
# 打印一步梯度下降后的loss
pred_t = linear(x_t)
loss_t = criterion(pred_t, y)
print('loss after 1 step optimization: ', loss_t.asnumpy().item())
# ================================================================== #
# 3. 从numpy中加载数据 #
# ================================================================== #
# 创建一个numpy数组
x_t = np.array([[1, 2], [3, 4]])
# 把numpy数组转化为tensor
y = mindspore.Tensor.from_numpy(x_t)
# 把tensor转化为numpy数组
z = y.asnumpy()
# ================================================================== #
# 4. 输入 #
# ================================================================== #
# 下载导入 CIFAR-10 数据集.
file_path = '../../../data/CIFAR-10'
if not os.path.exists(file_path):
if not os.path.exists('../../../data'):
os.mkdir('../../../data')
# 下载CIFAR-10数据集
os.mkdir(file_path)
url = 'https://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz'
file_name = 'cifar-10-binary.tar.gz'
print("正在从" + url + "下载CIFAR-10数据集...")
result = urllib.request.urlretrieve(url, os.path.join(file_path, file_name))
with tarfile.open(os.path.join(file_path, file_name), 'r:gz') as tar:
print("正在解压数据集...")
for member in tar.getmembers():
if member.name.startswith('cifar-10-batches-bin'):
member.name = os.path.basename(member.name)
tar.extract(member, path=file_path)
os.remove(os.path.join(file_path, file_name))
train_dataset = mindspore.dataset.Cifar10Dataset(
dataset_dir=file_path,
usage='train',
).map(operations=transforms.vision.ToTensor(), input_columns="image")
# 读取一组数据,创建并使用使用迭代器
for _, (image, label) in enumerate(train_dataset.create_tuple_iterator()):
print(ops.size(image))
print(label)
break
# ================================================================== #
# 5. 自定义数据集 #
# ================================================================== #
# 构建自定义数据集使用GeneratorDataset
# 详见https://www.mindspore.cn/docs/zh-CN/r2.0/api_python/dataset/mindspore.dataset.GeneratorDataset.html
# dataset = mindspore.dataset.GeneratorDataset(
# )
# ================================================================== #
# 6. 预训练模型 #
# ================================================================== #
# 从MindCV中加载Resnet18模型
resnet = mindcv.models.resnet18(pretrained=True)
# 如果你只想调整模型的顶层,可以按照以下方式设置
for param in resnet.trainable_params():
param.requires_grad = False
# 更换顶层进行微调
resnet.classifier = nn.Dense(resnet.classifier.in_channels, 100)
# 前向传播
images = ops.randn(64, 3, 224, 224)
outputs = resnet(images)
print(ops.shape(outputs)) # (64, 100)
# ================================================================== #
# 7. 保存和加载模型 #
# ================================================================== #
# Save and load the entire model.
mindspore.save_checkpoint(resnet, 'model.ckpt')
model = mindspore.load_checkpoint('model.ckpt')
# 保存优化器
mindspore.save_checkpoint(optimizer, 'optimizer.ckpt')
state_dict = mindspore.load_checkpoint('optimizer.ckpt')