JStarCraft RNS

git clone https://github.com/HongZhaoHua/jstarcraft-core.git

mvn install -Dmaven.test.skip=true

git clone https://github.com/HongZhaoHua/jstarcraft-ai.git

mvn install -Dmaven.test.skip=true

git clone https://github.com/HongZhaoHua/jstarcraft-rns.git

mvn install -Dmaven.test.skip=true

<dependency>
    <groupId>com.jstarcraft</groupId>
    <artifactId>rns</artifactId>
    <version>1.0</version>
</dependency>

compile group: 'com.jstarcraft', name: 'rns', version: '1.0'

Properties keyValues = new Properties();
keyValues.load(this.getClass().getResourceAsStream("/data.properties"));
keyValues.load(this.getClass().getResourceAsStream("/recommend/benchmark/randomguess-test.properties"));
Configurator configurator = new Configurator(keyValues);

RankingTask task = new RankingTask(RandomGuessModel.class, configurator);
// 训练与评估排序模型
task.execute();

RatingTask task = new RatingTask(RandomGuessModel.class, configurator);
// 训练与评估评分模型
task.execute();

// 获取模型
Model model = task.getModel();

随着信息技术和互联网的发展,人们逐渐从信息匮乏(Information Underload)的时代走入了信息过载(Information Overload)的时代.

无论是信息消费者还是信息生产者都遇到了挑战:
* 对于信息消费者,从海量信息中寻找信息,是一件非常困难的事情;
* 对于信息生产者,从海量信息中暴露信息,也是一件非常困难的事情;

信息检索的任务就是联系用户和信息,一方面帮助用户寻找对自己有价值的信息,另一方面帮助信息暴露给对它感兴趣的用户,从而实现信息消费者和信息生产者的双赢.

从信息检索的角度:
* 搜索和推荐是获取信息的两种主要手段;
* 搜索和推荐是获取信息的两种不同方式;
    * 搜索(Search)是主动明确的;
    * 推荐(Recommend)是被动模糊的;

搜索和推荐是两个互补的工具.

JStarCraft-RNS引擎旨在解决推荐与搜索领域的两个核心任务:排序预测(Ranking)和评分预测(Rating).

根据解决基本问题的不同,将算法与评估指标划分为排序(Ranking)与评分(Rating).

两者之间的根本区别在于目标函数的不同.
通俗点的解释:
Ranking算法基于隐式反馈数据,趋向于拟合用户的排序.(关注度)
Rating算法基于显示反馈数据,趋向于拟合用户的评分.(满意度)

关键在于具体场景中,关注度与满意度是否保持一致.
通俗点的解释:
人们关注的东西,并不一定是满意的东西.(例如:个人所得税)

// 构建配置
keyValues = new Properties();
keyValues.load(loader.getResourceAsStream("data.properties"));
keyValues.load(loader.getResourceAsStream("model/benchmark/randomguess-test.properties"));
configurator = new Configurator(keyValues);

// 此对象会返回给Java程序
_data = new HashMap();

// 构建排序任务
task = new RankingTask(RandomGuessModel.class, configurator);
// 训练与评估模型并获取排序指标
measures = task.execute();
_data.put("precision", measures.get(PrecisionEvaluator.class));
_data.put("recall", measures.get(RecallEvaluator.class));

// 构建评分任务
task = new RatingTask(RandomGuessModel.class, configurator);
// 训练与评估模型并获取评分指标
measures = task.execute();
_data.put("mae", measures.get(MAEEvaluator.class));
_data.put("mse", measures.get(MSEEvaluator.class));

_data;

 // 获取BeanShell脚本
File file = new File(ScriptTestCase.class.getResource("Model.bsh").toURI());
String script = FileUtils.readFileToString(file, StringUtility.CHARSET);

// 设置BeanShell脚本使用到的Java类
ScriptContext context = new ScriptContext();
context.useClasses(Properties.class, Assert.class);
context.useClass("Configurator", MapConfigurator.class);
context.useClasses("com.jstarcraft.ai.evaluate");
context.useClasses("com.jstarcraft.rns.task");
context.useClasses("com.jstarcraft.rns.model.benchmark");
// 设置BeanShell脚本使用到的Java变量
ScriptScope scope = new ScriptScope();
scope.createAttribute("loader", loader);

// 执行BeanShell脚本
ScriptExpression expression = new GroovyExpression(context, scope, script);
Map<String, Float> data = expression.doWith(Map.class);
Assert.assertEquals(0.005825241F, data.get("precision"), 0F);
Assert.assertEquals(0.011579763F, data.get("recall"), 0F);
Assert.assertEquals(1.2708743F, data.get("mae"), 0F);
Assert.assertEquals(2.425075F, data.get("mse"), 0F);

// 构建配置
def keyValues = new Properties();
keyValues.load(loader.getResourceAsStream("data.properties"));
keyValues.load(loader.getResourceAsStream("recommend/benchmark/randomguess-test.properties"));
def configurator = new Configurator(keyValues);

// 此对象会返回给Java程序
def _data = [:];

// 构建排序任务
task = new RankingTask(RandomGuessModel.class, configurator);
// 训练与评估模型并获取排序指标
measures = task.execute();
_data.precision = measures.get(PrecisionEvaluator.class);
_data.recall = measures.get(RecallEvaluator.class);

// 构建评分任务
task = new RatingTask(RandomGuessModel.class, configurator);
// 训练与评估模型并获取评分指标
measures = task.execute();
_data.mae = measures.get(MAEEvaluator.class);
_data.mse = measures.get(MSEEvaluator.class);

_data;

// 获取Groovy脚本
File file = new File(ScriptTestCase.class.getResource("Model.groovy").toURI());
String script = FileUtils.readFileToString(file, StringUtility.CHARSET);

// 设置Groovy脚本使用到的Java类
ScriptContext context = new ScriptContext();
context.useClasses(Properties.class, Assert.class);
context.useClass("Configurator", MapConfigurator.class);
context.useClasses("com.jstarcraft.ai.evaluate");
context.useClasses("com.jstarcraft.rns.task");
context.useClasses("com.jstarcraft.rns.model.benchmark");
// 设置Groovy脚本使用到的Java变量
ScriptScope scope = new ScriptScope();
scope.createAttribute("loader", loader);

// 执行Groovy脚本
ScriptExpression expression = new GroovyExpression(context, scope, script);
Map<String, Float> data = expression.doWith(Map.class);

// 构建配置
var keyValues = new Properties();
keyValues.load(loader.getResourceAsStream("data.properties"));
keyValues.load(loader.getResourceAsStream("recommend/benchmark/randomguess-test.properties"));
var configurator = new Configurator([keyValues]);

// 此对象会返回给Java程序
var _data = {};

// 构建排序任务
task = new RankingTask(RandomGuessModel.class, configurator);
// 训练与评估模型并获取排序指标
measures = task.execute();
_data['precision'] = measures.get(PrecisionEvaluator.class);
_data['recall'] = measures.get(RecallEvaluator.class);

// 构建评分任务
task = new RatingTask(RandomGuessModel.class, configurator);
// 训练与评估模型并获取评分指标
measures = task.execute();
_data['mae'] = measures.get(MAEEvaluator.class);
_data['mse'] = measures.get(MSEEvaluator.class);

_data;

// 获取JS脚本
File file = new File(ScriptTestCase.class.getResource("Model.js").toURI());
String script = FileUtils.readFileToString(file, StringUtility.CHARSET);

// 设置JS脚本使用到的Java类
ScriptContext context = new ScriptContext();
context.useClasses(Properties.class, Assert.class);
context.useClass("Configurator", MapConfigurator.class);
context.useClasses("com.jstarcraft.ai.evaluate");
context.useClasses("com.jstarcraft.rns.task");
context.useClasses("com.jstarcraft.rns.model.benchmark");
// 设置JS脚本使用到的Java变量
ScriptScope scope = new ScriptScope();
scope.createAttribute("loader", loader);

// 执行JS脚本
ScriptExpression expression = new JsExpression(context, scope, script);
Map<String, Float> data = expression.doWith(Map.class);

// 构建配置
var keyValues = Properties();
var loader = bindings["loader"] as ClassLoader;
keyValues.load(loader.getResourceAsStream("data.properties"));
keyValues.load(loader.getResourceAsStream("model/benchmark/randomguess-test.properties"));
var option = Option(keyValues);

// 此对象会返回给Java程序
var _data = mutableMapOf<String, Float>();

// 构建排序任务
var rankingTask = RankingTask(RandomGuessModel::class.java, option);
// 训练与评估模型并获取排序指标
val rankingMeasures = rankingTask.execute();
_data["precision"] = rankingMeasures.getFloat(PrecisionEvaluator::class.java);
_data["recall"] = rankingMeasures.getFloat(RecallEvaluator::class.java);

// 构建评分任务
var ratingTask = RatingTask(RandomGuessModel::class.java, option);
// 训练与评估模型并获取评分指标
var ratingMeasures = ratingTask.execute();
_data["mae"] = ratingMeasures.getFloat(MAEEvaluator::class.java);
_data["mse"] = ratingMeasures.getFloat(MSEEvaluator::class.java);

_data;

// 获取Kotlin脚本
File file = new File(ScriptTestCase.class.getResource("Model.kt").toURI());
String script = FileUtils.readFileToString(file, StringUtility.CHARSET);

// 设置Kotlin脚本使用到的Java类
ScriptContext context = new ScriptContext();
context.useClasses(Properties.class, Assert.class);
context.useClass("Option", MapOption.class);
context.useClasses("com.jstarcraft.ai.evaluate");
context.useClasses("com.jstarcraft.rns.task");
context.useClasses("com.jstarcraft.rns.model.benchmark");
// 设置Kotlin脚本使用到的Java变量
ScriptScope scope = new ScriptScope();
scope.createAttribute("loader", loader);

// 执行Kotlin脚本
ScriptExpression expression = new KotlinExpression(context, scope, script);
Map<String, Float> data = expression.doWith(Map.class);

-- 构建配置
local keyValues = Properties.new();
keyValues:load(loader:getResourceAsStream("data.properties"));

keyValues:load(loader:getResourceAsStream("recommend/benchmark/randomguess-test.properties"));
local configurator = Configurator.new({ keyValues });

-- 此对象会返回给Java程序
local _data = {};

-- 构建排序任务
task = RankingTask.new(RandomGuessModel, configurator);
-- 训练与评估模型并获取排序指标
measures = task:execute();
_data["precision"] = measures:get(PrecisionEvaluator);
_data["recall"] = measures:get(RecallEvaluator);

-- 构建评分任务
task = RatingTask.new(RandomGuessModel, configurator);
-- 训练与评估模型并获取评分指标
measures = task:execute();
_data["mae"] = measures:get(MAEEvaluator);
_data["mse"] = measures:get(MSEEvaluator);

return _data;

// 获取Lua脚本
File file = new File(ScriptTestCase.class.getResource("Model.lua").toURI());
String script = FileUtils.readFileToString(file, StringUtility.CHARSET);

// 设置Lua脚本使用到的Java类
ScriptContext context = new ScriptContext();
context.useClasses(Properties.class, Assert.class);
context.useClass("Configurator", MapConfigurator.class);
context.useClasses("com.jstarcraft.ai.evaluate");
context.useClasses("com.jstarcraft.rns.task");
context.useClasses("com.jstarcraft.rns.model.benchmark");
// 设置Lua脚本使用到的Java变量
ScriptScope scope = new ScriptScope();
scope.createAttribute("loader", loader);

// 执行Lua脚本
ScriptExpression expression = new LuaExpression(context, scope, script);
LuaTable data = expression.doWith(LuaTable.class);

# 构建配置
keyValues = Properties()
keyValues.load(loader.getResourceAsStream("data.properties"))
keyValues.load(loader.getResourceAsStream("recommend/benchmark/randomguess-test.properties"))
configurator = Configurator([keyValues])

# 此对象会返回给Java程序
_data = {}

# 构建排序任务
task = RankingTask(RandomGuessModel, configurator)
# 训练与评估模型并获取排序指标
measures = task.execute()
_data['precision'] = measures.get(PrecisionEvaluator)
_data['recall'] = measures.get(RecallEvaluator)

# 构建评分任务
task = RatingTask(RandomGuessModel, configurator)
# 训练与评估模型并获取评分指标
measures = task.execute()
_data['mae'] = measures.get(MAEEvaluator)
_data['mse'] = measures.get(MSEEvaluator)

// 设置Python环境变量
System.setProperty("python.console.encoding", StringUtility.CHARSET.name());

// 获取Python脚本
File file = new File(PythonTestCase.class.getResource("Model.py").toURI());
String script = FileUtils.readFileToString(file, StringUtility.CHARSET);

// 设置Python脚本使用到的Java类
ScriptContext context = new ScriptContext();
context.useClasses(Properties.class, Assert.class);
context.useClass("Configurator", MapConfigurator.class);
context.useClasses("com.jstarcraft.ai.evaluate");
context.useClasses("com.jstarcraft.rns.task");
context.useClasses("com.jstarcraft.rns.model.benchmark");
// 设置Python脚本使用到的Java变量
ScriptScope scope = new ScriptScope();
scope.createAttribute("loader", loader);

// 执行Python脚本
ScriptExpression expression = new PythonExpression(context, scope, script);
Map<String, Double> data = expression.doWith(Map.class);

# 构建配置
keyValues = Properties.new()
keyValues.load($loader.getResourceAsStream("data.properties"))
keyValues.load($loader.getResourceAsStream("model/benchmark/randomguess-test.properties"))
configurator = Configurator.new(keyValues)

# 此对象会返回给Java程序
_data = Hash.new()

# 构建排序任务
task = RankingTask.new(RandomGuessModel.java_class, configurator)
# 训练与评估模型并获取排序指标
measures = task.execute()
_data['precision'] = measures.get(PrecisionEvaluator.java_class)
_data['recall'] = measures.get(RecallEvaluator.java_class)

# 构建评分任务
task = RatingTask.new(RandomGuessModel.java_class, configurator)
# 训练与评估模型并获取评分指标
measures = task.execute()
_data['mae'] = measures.get(MAEEvaluator.java_class)
_data['mse'] = measures.get(MSEEvaluator.java_class)

_data;

// 获取Ruby脚本
File file = new File(ScriptTestCase.class.getResource("Model.rb").toURI());
String script = FileUtils.readFileToString(file, StringUtility.CHARSET);

// 设置Ruby脚本使用到的Java类
ScriptContext context = new ScriptContext();
context.useClasses(Properties.class, Assert.class);
context.useClass("Configurator", MapConfigurator.class);
context.useClasses("com.jstarcraft.ai.evaluate");
context.useClasses("com.jstarcraft.rns.task");
context.useClasses("com.jstarcraft.rns.model.benchmark");
// 设置Ruby脚本使用到的Java变量
ScriptScope scope = new ScriptScope();
scope.createAttribute("loader", loader);

// 执行Ruby脚本
ScriptExpression expression = new RubyExpression(context, scope, script);
Map<String, Double> data = expression.doWith(Map.class);
Assert.assertEquals(0.005825241096317768D, data.get("precision"), 0D);
Assert.assertEquals(0.011579763144254684D, data.get("recall"), 0D);
Assert.assertEquals(1.270874261856079D, data.get("mae"), 0D);
Assert.assertEquals(2.425075054168701D, data.get("mse"), 0D);