机器学习

Posted on Edited on In Word count in article: 623 Reading time ≈ 2 mins.

搭建私有仓库，上传 tensorflow 镜像

① 配置网络，能连通外网

② 安装docker

参考上篇的云容器平台实战，此处省略。。。

③ cc网下载 images.zip文件，上传 images文件（scp，WinSCP，XFTP，SecureCRT)

​ 先在物理机解压，再上传image文件 👇

​ scp -r images root@192.168.169.10:/root

④ 将registry_latest.tar映像导入 

1
2
3
[root@localhost ~]# docker load -i images/registry_latest.tar
查看registry映像
[root@localhost ~]# docker images

⑤创建一个 registry 容器

1
2
3
4
[root@localhost ~]# docker run -d --name registry -v /opt/registry:/var/lib/registry -p 5000:5000 --restart=always registry:latest

查看容器
[root@localhost ~]# docker ps

⑥ 配置仓库地址

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
[root@localhost ~]# vim /etc/docker/daemon.json
{
        "insecure-registries": ["192.168.169.10:5000"]		# 虚拟机IP:5000
}

重新加载模块、重启docker
[root@localhost ~]# systemctl daemon-reload
[root@localhost ~]# systemctl restart docker

访问测试
[root@localhost ~]# curl http://192.168.169.10:5000/v2
<a href="/v2/">Moved Permanently</a>.

[root@localhost ~]# curl http://192.168.169.10:5000/v2/_catalog
{"repositories":[]}

⑦ 上传 TensorFlow 镜像

1
2
3
4
5
6
7
8
9
10
11
12
13
14
下载tensorflow镜像
[root@localhost ~]# docker pull daocloud.io/daocloud/tensorflow:1.14.0-py3-jupyter

查看tensorflow镜像
[root@localhost ~]# docker images
REPOSITORY                        TAG                  IMAGE ID            CREATED             SIZE
daocloud.io/daocloud/tensorflow   1.14.0-py3-jupyter   fe3f97fcb6ce        3 years ago         1.34 GB
...

给tensorflow打标签
[root@localhost ~]# docker tag daocloud.io/daocloud/tensorflow:1.14.0-py3-jupyter 192.168.169.10:5000/tensorflow

上传镜像到私有仓库
[root@localhost ~]# docker push 192.168.169.10:5000/tensorflow

⑧ 访问测试

浏览器 http://192.168.169.10:5000/v2/_catalog

image-20220713004722801

识别手写数字

① 创建 tensorflow 容器

1
2
3
4
5
6
[root@localhost ~]# docker run -it --name my-tensorflow -p 8888:8888 -v ~/tensorflow:/test/data daocloud.io/daocloud/tensorflow:1.14.0-py3-jupyter 
...output omitted...
    Or copy and paste one of these URLs:
        http://(480dd667b304 or 127.0.0.1):8888/?token=a6a8f826b5fc2e2142bd2643930b68bebf6ae790e53cd1d8

访问 👉 http://'IP地址':8888/?token=a6a8f826b5fc2e2142bd2643930b68bebf6ae790e53cd1d8

![image-20220713093014367](Machine learning/image-20220713093014367.png)

② 上传mnist.npz和手写数字

image-20220713094229703

③ 编写模拟训练程序

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
import tensorflow as tf
import numpy as np

data=np.load('mnist.npz')

x_train,y_train,x_test,y_test=data['x_train'],data['y_train'],data['x_test'],data['y_test']

x_train = tf.keras.utils.normalize(x_train,axis=1)

x_test = tf.keras.utils.normalize(x_test,axis=1)

model = tf.keras.models.Sequential()

model.add(tf.keras.layers.Flatten(input_shape=(28,28)))
model.add(tf.keras.layers.Dense(999,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(999,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(10,activation=tf.nn.softmax))

model.compile(optimizer='adam',
             loss='sparse_categorical_crossentropy',
             metrics=['accuracy'])

model.fit(x_train,y_train,epochs=3)

val_loss,val_acc=model.evaluate(x_test,y_test)
print(val_loss)
print(val_acc)

model.save('my_model.h5')

model.summary()

④ 安装imageio

1
pip install --upgrade pip
1
pip install imageio -i https://pypi.douban.com/simple/

⑤ 导入测试图片

1
2
3
4
5
6
7
8
9
10
import glob
import imageio
import numpy as np
img_data = []

for image_file_name in glob.glob('img_?.jpg'):
        print("loading ...",image_file_name)
        img_array = imageio.imread(image_file_name,as_gray=True)
        img_data.append((255.0 - img_array.reshape(784)) / 255.0)
        self_data = np.array(img_data).reshape(len(img_data),28,28)

⑥ 进行预测

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
import numpy as np
import matplotlib.pyplot as plt
pre = model.predict(self_data)
predit=[0 for x in range(10)]
predit[0]=np.argmax(pre[9])
predit[1]=np.argmax(pre[8])
predit[2]=np.argmax(pre[7])
predit[3]=np.argmax(pre[6])
predit[4]=np.argmax(pre[5])
predit[5]=np.argmax(pre[4])
predit[6]=np.argmax(pre[3])
predit[7]=np.argmax(pre[2])
predit[8]=np.argmax(pre[1])
predit[9]=np.argmax(pre[0])

print(predit)