2 Tensorflow on Kubernetes

TensorFlow 是一个端到端开源机器学习平台。它拥有一个包含各种工具、库和社区资源的全面灵活生态系统，可以让研究人员推动机器学习领域的先进技术的发展，并让开发者轻松地构建和部署由机器学习提供支持的应用。

2.1 在 K8s 中安装 Tensorflow

使用 Webkubectl 进入集群 Terminal, 新建文件 tensorflow.yml 并输入以下内容:

---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: tensorflow-pvc
spec:
  accessModes:
  - ReadWriteOnce
  resources:
    requests:
      storage: 10Gi
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: tensorflow
spec:
  selector:
    matchLabels:
      k8s-app: tensorflow
  replicas: 1
  template:
    metadata:
      labels:
        k8s-app: tensorflow
    spec:
      containers:
      - name: tensorflow
        image: registry.cn-hangzhou.aliyuncs.com/tensorflow-samples/jupyter:1.1.0-devel-gpu
        imagePullPolicy: IfNotPresent
        env:
          - name: PASSWORD
            value: mypassw0rd
        resources:
          limits:
            nvidia.com/gpu: 1
        volumeMounts:
        - mountPath: /usr/local/nvidia
          name: nvidia
      volumes:
        - name: nvidia
          persistentVolumeClaim:
            claimName: tensorflow-pvc
---
apiVersion: v1
kind: Service
metadata:
  name: tensorflow-svc
spec:
  type: NodePort
  ports:
    - name: tensorflow-port
      port: 80
      targetPort: 8888
  selector:
     k8s-app: tensorflow

kubectl apply -f tensorflow.yml
persistentvolumeclaim/tensorflow-pvc configured
deployment.apps/tensorflow configured
service/tensorflow-svc configured
kubectl get pod
NAME                                   READY   STATUS    RESTARTS   AGE
tensorflow-79c5f4c48c-94l82            1/1     Running   0          101s
kubectl get svc
NAME             TYPE       CLUSTER-IP      EXTERNAL-IP   PORT(S)        AGE
tensorflow-svc   NodePort   179.10.100.67   <none>        80:32604/TCP   2m15s

访问 http://NODE_IP:32604 访问 Jupyter ，密码: mypassw0rd

2.2 运行 Tensorflow 深度学习示例

2.2.1 打开 Jupter 终端

2.2.2 使用 PIP 下载 KERAS

pip install keras==2.0.6

2.2.3 示例代码实现

import keras
from tensorflow.python.client import device_lib
# 获取 GPU 信息
#num_gpus = sum([1 for d in local_device_protos if d.device_type == 'GPU'])
#print("GPU : {}".format(num_gpus))
# 下载 MNIST 数据集
mnist = keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
# 将模型的各层堆叠起来，以搭建 keras.Sequential 模型。为训练选择优化器和损失函数：
model = keras.models.Sequential([
  keras.layers.Flatten(input_shape=(28, 28)),
  keras.layers.Dense(128, activation='relu'),
  keras.layers.Dropout(0.2),
  keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
# 训练并验证模型
model.fit(x_train, y_train, epochs=5)
model.evaluate(x_test,  y_test, verbose=2)

2.2.4 运行深度学习服务

python mnist.py
Epoch 1/5
60000/60000 [==============================] - 9s - loss: 0.2984 - acc: 0.9137     
Epoch 2/5
60000/60000 [==============================] - 7s - loss: 0.1422 - acc: 0.9575     
Epoch 3/5
60000/60000 [==============================] - 7s - loss: 0.1102 - acc: 0.9671     
Epoch 4/5
60000/60000 [==============================] - 7s - loss: 0.0885 - acc: 0.9728     
Epoch 5/5
60000/60000 [==============================] - 7s - loss: 0.0758 - acc: 0.9761 
[0.074264542010473084, 0.97750000000000004] #这个照片分类器的准确度已经达到 98%