烟草推荐模型部署文档

1、配置文件说明：

database_config.yaml 这个是数据配置文件

mysql:
host: 'rm-t4n6rz18y4t5x47y70o.mysql.singapore.rds.aliyuncs.com'
port: 3036
db: 'brand_cultivation'
user: 'xxxxx'
passwd: 'xxxxx'

redis:
host: 'r-t4nb4n9i8je7u6ogk1pd.redis.singapore.rds.aliyuncs.com'
port: 5000
db: 10
passwd: 'xxxxx'

crontab 定时任务配置文件

# START CRON JOB
1 2 * * * python /app/app.pyc --run_all
# END CRON JOB

2、模型启动配置说明：

app.py

    parser.add_argument("--run_all", action='store_true')
    parser.add_argument("--run_hot", action='store_true')
    parser.add_argument("--run_itemcf", action='store_true')
    parser.add_argument("--run_itemcf_inference", action='store_true')
    parser.add_argument("--city_uuid", type=str, help="City UUID for filtering data")

总共有4种启动模式分别是：

1. 启动热度召回和协同过滤
2. 启动热度召回
3. 启动协同过滤
4. 启动系统过滤推理

3、GBDT LR模型训练推理启动

gbdt_lr.py

    parser.add_argument("--run_train", action='store_true')
    parser.add_argument("--recommend", action='store_true')
    parser.add_argument("--importance", action='store_true')

    parser.add_argument("--city_uuid", type=str, default='00000000000000000000000011445301')
    parser.add_argument("--product_id", type=str, default='110102')

gbdt_lr总共3个功能：

1. 启动gbdt_lr训练 python -m gbdt_lr --run_train --city_uuid "00000000000000000000000011445301"
2. 根据城市id和product_id进行推荐，需要指定city_uuid、product_id。 python -m gbdt_lr --recommend --city_uuid "00000000000000000000000011445301" --product_id '110102'
3. 获取指定城市的特征重要性指标。 python -m gbdt_lr --importance --city_uuid "00000000000000000000000011445301"
注意：在数据准备阶段，会将训练数据保存到./models/rank/data/gbdt_data.csv中
模型文件会存放在 ./models/rank/weights/city_uuid/model.pkl
重要性指标会存放在 ./models/rank/weights/下，分别是商户指标重要性和卷烟指标重要性

4、模型docker运行配置说明：

docker镜像是：registry.cn-hangzhou.aliyuncs.com/hexiaoshi/brandcultivation:0.0.1

docker run --name BrandCultivation -d -v /export/brandcultivation/crontab:/etc/cron.d/crontab -v /export/brandcultivation/database_config.yaml:/app/config/database_config.yaml  registry.cn-hangzhou.aliyuncs.com/hexiaoshi/brandcultivation:0.0.1

5、模型kubernetes运行配置说明

yaml文件如下：

apiVersion: apps/v1
kind: Deployment
metadata:
  name: brandcultivation
  namespace: default
  labels:
    app: brandcultivation
spec:
  selector:
    matchLabels:
      app: brandcultivation
  replicas: 1
  strategy:
    rollingUpdate:
      maxSurge: 25%
      maxUnavailable: 25%
    type: RollingUpdate
  template:
    metadata:
      labels:
        app: brandcultivation
    spec:
      containers:
      - name: brandcultivation
        image: registry.cn-hangzhou.aliyuncs.com/hexiaoshi/brandcultivation:0.0.1
        imagePullPolicy: IfNotPresent
        resources:
          requests:
            cpu: 4000m
            memory: 4096Mi
            ephemeral-storage: 20Gi             
          limits:
            cpu: 4000m
            memory: 4096Mi
            ephemeral-storage: 20Gi            
        ports:
        - containerPort:  80
          name: brandcultivation
        volumeMounts:
        - name: localtime
          mountPath: /etc/localtime
        - name: config
          mountPath: /app/config/database_config.yaml
          subPath: database_config.yaml
        - name: config
          mountPath: /etc/cron.d/crontab
          subPath: crontab          
        - name: localtime
          hostPath:
            path: /usr/share/zoneinfo/Asia/Shanghai
        - name: config
          configMap:
            name: brandcultivation
      restartPolicy: Always
---
kind: ConfigMap
apiVersion: v1
metadata:
  name: brandcultivation
  namespace: default
data:
  database_config.yaml: |
    mysql:
      host: 'rm-t4n6rz18y4t5x47y70o.mysql.singapore.rds.aliyuncs.com'
      port: 3036
      db: 'brand_cultivation'
      user: 'BrandCultivation'
      passwd: '8BfWBc18NBXl#CMd'

    redis:
      host: 'r-t4nb4n9i8je7u6ogk1pd.redis.singapore.rds.aliyuncs.com'
      port: 5000
      db: 10
      passwd: 'gHmNkVBd88sZybj'
  crontab: |
    # START CRON JOB
    1 2 * * * python /app/app.pyc
    # END CRON JOB

烟草模型部署文档.md 4.6 KB Historia Czysty