基于Azure构建PredictionIO和Spark的推荐引擎服务

1. 在Azure构建Ubuntu 16.04虚拟机

假设前提条件您已有 Azure 帐号,登陆 Azure https://portal.azure.com

点击左上部的 +New 按钮,在搜索框中输入 Ubuntu ,或者点击 Virtual Machine 选择 Ubuntu Server 14.04 LTS,然后点击 Create 创建虚拟机。

创建完成虚拟机后,在虚拟机的设置中找到 Azure 为其分配的 IP 地址,通过 Bitvise SSH Client 远程登陆虚拟机开始快速搭建推荐引擎服务之旅。

2. 以 PredictionIO 源码编译的方式安装

之所以本文以源码编译的方式安装 PredictionIO ,是因为其他方式我都未尝试成功。

2.1 编译

Run the following to download and build Apache PredictionIO (incubating) from its source code.

$ git clone https://github.com/apache/incubator-predictionio.git
$ cd incubator-predictionio
$ git checkout master
$ ./make-distribution.sh

You should see something like the following when it finishes building successfully.

... PredictionIO-0.9.6/sbt/sbt PredictionIO-0.9.6/conf/

PredictionIO-0.9.6/conf/pio-env.sh PredictionIO binary distribution

created at PredictionIO-0.9.6.tar.gz Extract the binary distribution

you have just built.

$ tar zxvf PredictionIO-0.9.6.tar.gz

2.2 安装依赖

Let us install dependencies inside a subdirectory of the Apache PredictionIO (incubating) installation. By following this convention, you can use Apache PredictionIO (incubating)'s default configuration as is.

$ mkdir PredictionIO-0.9.6/vendors

2.3 安装Spark依赖包

Apache Spark is the default processing engine for PredictionIO. Download and extract it.

$ wget http://d3kbcqa49mib13.cloudfront.net/spark-1.5.1-bin-hadoop2.6.tgz
$ tar zxvfC spark-1.5.1-bin-hadoop2.6.tgz PredictionIO-0.9.6/vendors

If you decide to install Apache Spark to another location, you must edit PredictionIO-0.9.6/conf/pio-env.sh and change the SPARK_HOME variable to point to your own Apache Spark installation.

2.4 数据存储

官方给的例子是采用 PostgreSQL 或者Hbase + Elasticsearch,我选择 MySQL 作为数据存储,因为在将来的数据可视化方面会采用 Caravel 自动化生成仪表板展现数据,在后续的文章中我会再详细介绍这方面。

在mysql官方网站下载 mysql-connector-java-5.1.37.jar 并保存至 PredictionIO-0.9.6/lib 文件夹中。

修改pi配置文件 pio-env.sh

#!/usr/bin/env bash 

# Copy this file as pio-env.sh and edit it for your site's configuration. 

# PredictionIO Main Configuration
#
# This section controls core behavior of PredictionIO. It is very likely that
# you need to change these to fit your site. # SPARK_HOME: Apache Spark is a hard dependency and must be configured.
SPARK_HOME=$PIO_HOME/vendors/spark-1.5.1-bin-hadoop2.6 #POSTGRES_JDBC_DRIVER=$PIO_HOME/lib/postgresql-9.4-1204.jdbc41.jar
MYSQL_JDBC_DRIVER=$PIO_HOME/lib/mysql-connector-java-5.1.37.jar # ES_CONF_DIR: You must configure this if you have advanced configuration for
# your Elasticsearch setup.
# ES_CONF_DIR=/opt/elasticsearch # HADOOP_CONF_DIR: You must configure this if you intend to run PredictionIO
# with Hadoop 2.
# HADOOP_CONF_DIR=/opt/hadoop # HBASE_CONF_DIR: You must configure this if you intend to run PredictionIO
# with HBase on a remote cluster.
# HBASE_CONF_DIR=$PIO_HOME/vendors/hbase-1.0.0/conf # Filesystem paths where PredictionIO uses as block storage.
PIO_FS_BASEDIR=$HOME/.pio_store
PIO_FS_ENGINESDIR=$PIO_FS_BASEDIR/engines
PIO_FS_TMPDIR=$PIO_FS_BASEDIR/tmp # PredictionIO Storage Configuration
#
# This section controls programs that make use of PredictionIO's built-in
# storage facilities. Default values are shown below.
#
# For more information on storage configuration please refer to
# https://docs.prediction.io/system/anotherd
# Storage Repositories # Default is to use PostgreSQL
PIO_STORAGE_REPOSITORIES_METADATA_NAME=pio_meta
PIO_STORAGE_REPOSITORIES_METADATA_SOURCE=MYSQL PIO_STORAGE_REPOSITORIES_EVENTDATA_NAME=pio_event
PIO_STORAGE_REPOSITORIES_EVENTDATA_SOURCE=MYSQL PIO_STORAGE_REPOSITORIES_MODELDATA_NAME=pio_model
PIO_STORAGE_REPOSITORIES_MODELDATA_SOURCE=MYSQL # Storage Data Sources # PostgreSQL Default Settings
# Please change "pio" to your database name in PIO_STORAGE_SOURCES_PGSQL_URL
# Please change PIO_STORAGE_SOURCES_PGSQL_USERNAME and
# PIO_STORAGE_SOURCES_PGSQL_PASSWORD accordingly
#PIO_STORAGE_SOURCES_PGSQL_TYPE=jdbc
#PIO_STORAGE_SOURCES_PGSQL_URL=jdbc:postgresql://localhost/pio
#PIO_STORAGE_SOURCES_PGSQL_USERNAME=pio
#PIO_STORAGE_SOURCES_PGSQL_PASSWORD=pio # MySQL Example
PIO_STORAGE_SOURCES_MYSQL_TYPE=jdbc
PIO_STORAGE_SOURCES_MYSQL_URL=jdbc:mysql://10.18.218.9:13306/xuesongpio
PIO_STORAGE_SOURCES_MYSQL_USERNAME=root
PIO_STORAGE_SOURCES_MYSQL_PASSWORD=****** # Elasticsearch Example
# PIO_STORAGE_SOURCES_ELASTICSEARCH_TYPE=elasticsearch
# PIO_STORAGE_SOURCES_ELASTICSEARCH_CLUSTERNAME=<elasticsearch_cluster_name>
# PIO_STORAGE_SOURCES_ELASTICSEARCH_HOSTS=localhost
# PIO_STORAGE_SOURCES_ELASTICSEARCH_PORTS=9300
# PIO_STORAGE_SOURCES_ELASTICSEARCH_HOME=$PIO_HOME/vendors/elasticsearch-1.4.4 # Local File System Example
# PIO_STORAGE_SOURCES_LOCALFS_TYPE=localfs
# PIO_STORAGE_SOURCES_LOCALFS_PATH=$PIO_FS_BASEDIR/models
atastore/ # HBase Example
# PIO_STORAGE_SOURCES_HBASE_TYPE=hbase
# PIO_STORAGE_SOURCES_HBASE_HOME=$PIO_HOME/vendors/hbase-1.0.0

2.5声明 PrecidtionIO 和 Java 的环境变量

我是在 /root/.bahsrc 里增加如下代码:

PIO_HOME=/root/PredictionIO-0.9.6
JAVA_HOME=/usr/lib/jvm/java-8-oracle
JRE_HOME=$JAVA_HOME/jre
PATH=$PATH:$PIO_HOME/bin:$JAVA_HOME/bin:$JRE_HOME/bin
export PIO_HOME
export JAVA_HOME
export PATH

3. 创建推荐引擎服务

3.1 启动 PredictionIO Event Server

因为我在运行 PredictionIO 过程中发现它还是挺吃内存的,不管是 PredictionIO 还是 Spark,因此我特别分配的16的内存运行 Event server

JAVA_OPTS=-Xmx16g bin/pio eventserver &

3.2 启动 Spark

通过 PredictionIO 内置的脚本启动 Spark 进行模型训练时总是出现问题,所以采用手动启动 Spark 集群的方式规避此问题。

You can start a standalone master server by executing:

PredictionIO-0.9.6/vendors/spark-1.5.1-bin-hadoop2.6/sbin/start-master.sh

Once started, the master will print out a spark://HOST:PORT URL for itself, which you can use to connect workers to it, or pass as the "master" argument to SparkContext. You can also find this URL on the master's web UI, which is http://localhost:8080 by default.

Similarly, you can start one or more workers and connect them to the master via:

./sbin/start-slave.sh <master-spark-URL>

PredictionIO-0.9.6/vendors/spark-1.5.1-bin-hadoop2.6/sbin/start-slave.sh spark://localhost:8080

Once you have started a worker, look at the master's web UI (http://localhost:8080 by default). You should see the new node listed there, along with its number of CPUs and memory (minus one gigabyte left for the OS).

3.3 Create a new Engine from an Engine Template

Now let's create a new engine called MyRecommendation by downloading the Recommendation Engine Template. Go to a directory where you want to put your engine and run the following:

$ pio template get PredictionIO/template-scala-parallel-recommendation MyRecommendation
$ cd MyRecommendation

A new directory MyRecommendation is created, where you can find the downloaded engine template.

3.4 Generate an App ID and Access Key

You will need to create a new App in PredictionIO to store all the data of your app. The data collected will be used for machine learning modeling.

Let's assume you want to use this engine in an application named "MyApp1". Run the following to create a new app "MyApp1":

$ pio app new MyApp1

You should find the following in the console output:

... [INFO] [App\(] Initialized Event Store for this app ID: 1. [INFO]
[App\)] Created new app: [INFO] [App\(] Name: MyApp1 [INFO] [App\)] ID: 1

[INFO] [App$] Access Key:

3mZWDzci2D5YsqAnqNnXH9SB6Rg3dsTBs8iHkK6X2i54IQsIZI1eEeQQyMfs7b3F Note

that App ID, *Access Key are created for this App "MyApp1". You will need the Access Key when you collect data with EventServer for this App.

You can list all of the apps created its corresponding ID and Access Key by running the following command:

$ pio app list

You should see a list of apps created. For example:

[INFO] [App\(] Name | ID | Access Key | Allowed Event(s) [INFO] [App\)]

MyApp1 | 1 |

3mZWDzci2D5YsqAnqNnXH9SB6Rg3dsTBs8iHkK6X2i54IQsIZI1eEeQQyMfs7b3F |

(all) [INFO] [App\(] MyApp2 | 2 |
io5lz6Eg4m3Xe4JZTBFE13GMAf1dhFl6ZteuJfrO84XpdOz9wRCrDU44EUaYuXq5 |
(all) [INFO] [App\)] Finished listing 2 app(s).

3.5 Import More Sample Data

This engine requires more data in order to train a useful model. Instead of sending more events one by one in real time, for quickstart demonstration purpose, we are going to use a script to import more events in batch.

A Python import script import_eventserver.py is provided in the template to import the data to Event Server using Python SDK. Please upgrade to the latest Python SDK.

First, you will need to install Python SDK in order to run the sample data import script. To install Python SDK, run:

$ pip install predictionio

You may need sudo access if you have permission issue. (ie. sudo pip install predictionio)

Replace the value of access_key parameter by your Access Key and run:

These commands must be executed in the Engine directory, for example: MyRecomendation.

$ cd MyRecommendation
$ curl https://raw.githubusercontent.com/apache/spark/master/data/mllib/sample_movielens_data.txt --create-dirs -o data/sample_movielens_data.txt
$ python data/import_eventserver.py --access_key $ACCESS_KEY

You should see the following output:

Importing data... 1501 events are imported.

3.6 编译模型

Start with building your MyRecommendation engine. Run the following command:

$ pio build --verbose

This command should take few minutes for the first time; all subsequent builds should be less than a minute. You can also run it without --verbose if you don't want to see all the log messages.

Upon successful build, you should see a console message similar to the following.

[INFO] [Console$] Your engine is ready for training.

3.7 训练模型

在上文中提到 Spark 集群是由我们手工启动的,因此在模型训练时指定 Spark master UI 和运行时内存参数。

To train your engine, run the following command:

pio train -- --master spark://localhost:7077 --driver-memory 16G --executor-memory 24G

Deploy Engine

To increase the heap space, specify the "-- --driver-memory " parameter in the command. For example, set the driver memory to 8G when deploy the engine:

$ pio deploy -- --driver-memory 8G

When the engine is deployed successfully and running, you should see a console message similar to the following:

[INFO] [HttpListener] Bound to /0.0.0.0:8000 [INFO] [MasterActor] Bind

successful. Ready to serve.

Do not kill the deployed engine process.

By default, the deployed engine binds to http://localhost:8000. You can visit that page in your web browser to check its status.

3.8 Use the Engine

Now, You can try to retrieve predicted results. To recommend 4 movies to user whose id is 1, you send this JSON { "user": "1", "num": 4 } to the deployed engine and it will return a JSON of the recommended movies. Simply send a query by making a HTTP request or through the EngineClient of an SDK.

With the deployed engine running, open another temrinal and run the following curl command or use SDK to send the query:

$ curl -H "Content-Type: application/json" \
-d '{ "user": "1", "num": 4 }' http://localhost:8000/queries.json

或者执行 Python 程序,程序代码如下:

import predictionio
engine_client = predictionio.EngineClient(url="http://localhost:8000")
print engine_client.send_query({"user": "1", "num": 4})

The following is sample JSON response:

{
"itemScores":[
{"item":"22","score":4.072304374729956},
{"item":"62","score":4.058482414005789},
{"item":"75","score":4.046063009943821},
{"item":"68","score":3.8153661512945325}
]
}

Congratulations, MyRecommendation is now running!

本文被批评文档格式不好,特意用 markdown 重新编辑一遍,请笑纳。

基于Azure构建PredictionIO和Spark的推荐引擎服务的更多相关文章

  1. Azure构建PredictionIO和Spark的推荐引擎服务

    Azure构建PredictionIO和Spark的推荐引擎服务 1. 在Azure构建Ubuntu 16.04虚拟机 假设前提条件您已有 Azure 帐号,登陆 Azure https://port ...

  2. Spark学习笔记——构建基于Spark的推荐引擎

    推荐模型 推荐模型的种类分为: 1.基于内容的过滤:基于内容的过滤利用物品的内容或是属性信息以及某些相似度定义,来求出与该物品类似的物品. 2.协同过滤:协同过滤是一种借助众包智慧的途径.它利用大量已 ...

  3. PredictionIO+Universal Recommender快速开发部署推荐引擎的问题总结(3)

    PredictionIO+Universal Recommender虽然可以帮助中小企业快速的搭建部署基于用户行为协同过滤的个性化推荐引擎,单纯从引擎层面来看,开发成本近乎于零,但仍然需要一些前提条件 ...

  4. 基于Spark ALS构建商品推荐引擎

    基于Spark ALS构建商品推荐引擎   一般来讲,推荐引擎试图对用户与某类物品之间的联系建模,其想法是预测人们可能喜好的物品并通过探索物品之间的联系来辅助这个过程,让用户能更快速.更准确的获得所需 ...

  5. 【Microsoft Azure 的1024种玩法】二.基于Azure云平台的安全攻防靶场系统构建

    简介 本篇文章将基于在Microsoft Azure云平台上使用Pikachu去构建安全攻防靶场,Pikachu使用世界上最好的语言PHP进行开发,数据库使用的是mysql,因此运行Pikachu需要 ...

  6. [转] 基于 Apache Mahout 构建社会化推荐引擎

    来源:http://www.ibm.com/developerworks/cn/java/j-lo-mahout/index.html 推荐引擎简介 推荐引擎利用特殊的信息过滤(IF,Informat ...

  7. 基于 Apache Mahout 构建社会化推荐引擎

    基于 Apache Mahout 构建社会化推荐引擎 http://www.ibm.com/developerworks/cn/views/java/libraryview.jsp 推荐引擎利用特殊的 ...

  8. 【Microsoft Azure 的1024种玩法】三.基于Azure云平台构建Discuz论坛

    [简介] Discuz!是一套通用社区论坛软件系统,用户在不需要任何编程的基础上,通过简单的设置和安装,在互联网上搭建起具备完善功能.很强负载能力和可高度定制的论坛服务. [前期文章] [操作步骤] ...

  9. 基于Kubernetes 构建.NET Core技术中台

    今天下午在腾讯云+社区社区分享了<基于Kubernetes 构建.NET Core技术中台>,下面是演讲内容的文字实录. 我们为什么需要中台 我们现在处于企业信息化的新时代.为什么这样说呢 ...

随机推荐

  1. [Q&A] 类Range的PasteSpecial方法无效

    环境说明: VS2013(C#) + Office2013 Bug说明: range1.Copy(Type.Missing); range2.PasteSpecial(Excel.XlPasteTyp ...

  2. Artificial intelligence(AI)

    ORM: https://github.com/sunkaixuan/SqlSugar 微软DEMO: https://github.com/Microsoft/BotBuilder 注册KEY:ht ...

  3. swift相关

    1.界面上一个输入框,一个按钮,一个webview .点按钮,webview显示输入框中输入的url的页面. @IBOutlet var wv:UIWebView = nil          //定 ...

  4. Mac入门(一)基本用法

    我前五年一直外包到微软,每天使用的都是Windows系统和.NET. 2012年加入VMware,  公司的工作机是台Mac 笔记本(MacBook Pro), 所以有机会接触Mac系统 Mac和Wi ...

  5. Lind.DDD.Repositories.EF层介绍

    回到目录 Lind.DDD.Repositories.EF以下简称Repositories.EF,之所以把它从Lind.DDD中拿出来,完全出于可插拔的考虑,让大家都能休会到IoC的魅力,用到哪种方法 ...

  6. java web学习总结(八) -------------------HttpServletResponse对象(二)

    一.HttpServletResponse常见应用--生成验证码 1.1.生成随机图片用作验证码 生成图片主要用到了一个BufferedImage类,

  7. js中的位运算

    按位运算符是把操作数看作一系列单独的位,而不是一个数字值.所以在这之前,不得不提到什么是"位": 数值或字符在内存内都是被存储为0和 1的序列,每个0和1被称之为1个位,比如说10 ...

  8. [deviceone开发]-小草用户分享的Listview停靠的示例

    一.简介 这个例子展示了Listview的多模板,上拉下拉功能,也实现了上下滑动第二行工具栏的停靠功能,值得参考 二.效果图 三.相关下载 https://github.com/do-project/ ...

  9. Python开发【第三篇】:Python基本之文件操作

    Python基本之文本操作 一.初识文本的基本操作 在python中打开文件有两种方式,即:open(...) 和  file(...) ,本质上前者在内部会调用后者来进行文件操作,推荐使用 open ...

  10. IE7浏览器窗口大小改变事件执行多次bug(转)

    var resizeTimer = null; $(window).resize(function() { if (resizeTimer) clearTimeout(resizeTimer); re ...