|
|
@@ -0,0 +1,198 @@
|
|
|
+package com.aliyun.odps.spark.examples.makedata
|
|
|
+
|
|
|
+import com.alibaba.fastjson.{JSON, JSONObject}
|
|
|
+import com.aliyun.odps.TableSchema
|
|
|
+import com.aliyun.odps.data.Record
|
|
|
+import com.aliyun.odps.spark.examples.myUtils.{MyDateUtils, MyHdfsUtils, ParamUtils}
|
|
|
+import examples.dataloader.{OfflineVlogShareLRFeatureExtractorV1, OfflineVlogShareLRFeatureExtractorV2}
|
|
|
+import org.apache.hadoop.io.compress.GzipCodec
|
|
|
+import org.apache.spark.sql.SparkSession
|
|
|
+
|
|
|
+import java.util
|
|
|
+import scala.collection.JavaConversions._
|
|
|
+import scala.collection.mutable.ArrayBuffer
|
|
|
+import examples.extractor.ExtractorUtils
|
|
|
+
|
|
|
+object makedata_12_rosData_v3 {
|
|
|
+ def main(args: Array[String]) {
|
|
|
+ val spark = SparkSession
|
|
|
+ .builder()
|
|
|
+ .appName(this.getClass.getName)
|
|
|
+ .getOrCreate()
|
|
|
+ val sc = spark.sparkContext
|
|
|
+
|
|
|
+ // 1 读取参数
|
|
|
+ val param = ParamUtils.parseArgs(args)
|
|
|
+ val partitionPrefix = param.getOrElse("partitionPrefix", "dt=")
|
|
|
+ val beginStr = param.getOrElse("beginStr", "20230101")
|
|
|
+ val endStr = param.getOrElse("endStr", "20230101")
|
|
|
+ val readPath = param.getOrElse("readPath", "/dw/recommend/model/10_sample_data_v3/")
|
|
|
+ val savePath = param.getOrElse("savePath", "/dw/recommend/model/12_ros_data_v3/")
|
|
|
+ val featureVersion = param.getOrElse("featureVersion", "v2")
|
|
|
+ val ifRepart = param.getOrElse("ifRepart", "10").toInt
|
|
|
+ val labelVersion = param.getOrElse("labelVersion", "v1")
|
|
|
+
|
|
|
+ // 3 循环执行数据生产
|
|
|
+ val dateRange = MyDateUtils.getDateRange(beginStr, endStr)
|
|
|
+ for (date <- dateRange) {
|
|
|
+ val partition = partitionPrefix + date
|
|
|
+ println("开始执行partiton:" + partition)
|
|
|
+ var hdfsPath = readPath + "/" + partition
|
|
|
+
|
|
|
+ //1 样本过滤(分享样本、012345中的、可推荐的video、不同产品)
|
|
|
+ val data1 = sc.textFile(hdfsPath).map(r => {
|
|
|
+ val rList = r.split("\t")
|
|
|
+ val logKeyStr = rList(0)
|
|
|
+ val (mid, videoid, logtimestamp, apptype, pagesource_change, abcode, video_recommend) = ParamUtils.parseLogKey(logKeyStr)
|
|
|
+ val labelStr = rList(1)
|
|
|
+ val feaStr = rList(2)
|
|
|
+ val labelJson = JSON.parseObject(labelStr)
|
|
|
+ val is_share = labelJson.getString("is_share")
|
|
|
+ (logKeyStr, labelJson, feaStr, is_share, pagesource_change, video_recommend, apptype, logtimestamp.toLong)
|
|
|
+ }).filter({
|
|
|
+ case (logKeyStr, labelJson, feaStr, is_share, pagesource_change, video_recommend, apptype, logtimestamp) =>
|
|
|
+ val pages = Set("2")
|
|
|
+ val video_status = Set("-6")
|
|
|
+ val apps = Set("0", "4", "5", "21", "3", "6")
|
|
|
+ "1".equals(is_share) && pages.contains(pagesource_change) && video_status.contains(video_recommend) && apps.contains(apptype)
|
|
|
+ })
|
|
|
+
|
|
|
+ //2 样本采样（多个回流的样本复制，等价回流量的加权）
|
|
|
+ val data2 = data1.flatMap({
|
|
|
+ case (logKeyStr, labelJson, feaStr, is_share, pagesource_change, video_recommend, apptype, logtimestamp) =>
|
|
|
+ val res = ArrayBuffer[(String, JSONObject)]()
|
|
|
+ val feaJson = JSON.parseObject(feaStr)
|
|
|
+ val is_return = labelJson.getString("is_return")
|
|
|
+ if ("0".equals(is_return)){
|
|
|
+ res.add(("0", feaJson))
|
|
|
+ }else{
|
|
|
+ val return_mid_ts_list = labelJson.getString("return_mid_ts_list").split(",").map(r => {
|
|
|
+ val midReturn = r.split(":")(0)
|
|
|
+ val ts = r.split(":")(1).toLong / 1000
|
|
|
+ (midReturn, ts)
|
|
|
+ }).sortBy(_._2)
|
|
|
+ val midSet = Set()
|
|
|
+ for ((midReturn, tsReturn) <- return_mid_ts_list){
|
|
|
+ if (!midSet.contains(midReturn)){
|
|
|
+ midSet.add(midReturn)
|
|
|
+ if (tsReturn - logtimestamp <= 3600 && tsReturn - logtimestamp > 0){
|
|
|
+ res.add(("1", feaJson))
|
|
|
+ }
|
|
|
+ }
|
|
|
+ }
|
|
|
+ }
|
|
|
+ res.iterator
|
|
|
+ })
|
|
|
+
|
|
|
+ //3 保留一份原始样本的中间数据
|
|
|
+ println("样本比例")
|
|
|
+ data2.map(r=> (r._1, 1)).reduceByKey(_+_).map(r=> r._1 + "\t" + r._2).collect().foreach(println)
|
|
|
+
|
|
|
+ //4 特征绝对值转换 如 0.456变成19
|
|
|
+ val data3 = data2.map({
|
|
|
+ case (label, feaJson) =>
|
|
|
+ Set(
|
|
|
+ "u_1day_ctr", "u_1day_str", "u_1day_rov", "u_1day_ros",
|
|
|
+ "u_3day_ctr", "u_3day_str", "u_3day_rov", "u_3day_ros",
|
|
|
+ "u_7day_ctr", "u_7day_str", "u_7day_rov", "u_7day_ros",
|
|
|
+ "u_3month_ctr", "u_3month_str", "u_3month_rov", "u_3month_ros",
|
|
|
+ // ----------
|
|
|
+ "i_1day_ctr", "i_1day_str", "i_1day_rov", "i_1day_ros",
|
|
|
+ "i_3day_ctr", "i_3day_str", "i_3day_rov", "i_3day_ros",
|
|
|
+ "i_7day_ctr", "i_7day_str", "i_7day_rov", "i_7day_ros",
|
|
|
+ "i_3month_ctr", "i_3month_str", "i_3month_rov", "i_3month_ros",
|
|
|
+ // ----------
|
|
|
+ "i_1day_ctr_rt", "i_1day_str_rt", "i_1day_ros_rt", "i_1day_rov_rt",
|
|
|
+ "i_1h_ctr_rt", "i_1h_str_rt", "i_1h_ros_rt", "i_1h_rov_rt"
|
|
|
+ ).foreach(key =>{
|
|
|
+ if (feaJson.containsKey(key)){
|
|
|
+ val value = ExtractorUtils.ceilLogRate(feaJson.getString(key).toDouble)
|
|
|
+ feaJson.put(key, value.toString)
|
|
|
+ }
|
|
|
+ })
|
|
|
+ Set(
|
|
|
+ "u_1day_exp_cnt", "u_1day_click_cnt", "u_1day_share_cnt", "u_1day_return_cnt",
|
|
|
+ "u_3day_exp_cnt", "u_3day_click_cnt", "u_3day_share_cnt", "u_3day_return_cnt",
|
|
|
+ "u_7day_exp_cnt", "u_7day_click_cnt", "u_7day_share_cnt", "u_7day_return_cnt",
|
|
|
+ "u_3month_exp_cnt", "u_3month_click_cnt", "u_3month_share_cnt", "u_3month_return_cnt",
|
|
|
+ // ----------
|
|
|
+ "total_time", "play_count", "play_count_total",
|
|
|
+ "i_1day_exp_cnt", "i_1day_click_cnt", "i_1day_share_cnt", "i_1day_return_cnt",
|
|
|
+ "i_3day_exp_cnt", "i_3day_click_cnt", "i_3day_share_cnt", "i_3day_return_cnt",
|
|
|
+ "i_7day_exp_cnt", "i_7day_click_cnt", "i_7day_share_cnt", "i_7day_return_cnt",
|
|
|
+ "i_3month_exp_cnt", "i_3month_click_cnt", "i_3month_share_cnt", "i_3month_return_cnt",
|
|
|
+ // ----------
|
|
|
+ "share_uv_list_1day_6_avg", "share_uv_list_1day_6_var", "share_uv_list_1day_diff_6_avg", "share_uv_list_1day_diff_6_var",
|
|
|
+ "return_uv_list_1day_6_avg", "return_uv_list_1day_6_var", "return_uv_list_1day_diff_6_avg", "return_uv_list_1day_diff_6_var",
|
|
|
+ "share_uv_list_1h_6_avg", "share_uv_list_1h_6_var", "share_uv_list_1h_diff_6_avg", "share_uv_list_1h_diff_6_var",
|
|
|
+ "return_uv_list_1h_6_avg", "return_uv_list_1h_6_var", "return_uv_list_1h_diff_6_avg", "return_uv_list_1h_diff_6_var",
|
|
|
+ // ----------
|
|
|
+ "view_pv_list_1day", "view_uv_list_1day", "play_pv_list_1day", "play_uv_list_1day",
|
|
|
+ "share_pv_list_1day", "share_uv_list_1day", "return_uv_list_1day",
|
|
|
+ "p_view_uv_list_1day", "p_view_pv_list_1day", "p_return_uv_list_1day",
|
|
|
+ "share_uv_list_2day", "share_pv_list_2day", "share_uv_list_3day", "share_pv_list_3day",
|
|
|
+ // ----------
|
|
|
+ "view_uv_list_1h", "view_pv_list_1h", "play_uv_list_1h", "play_pv_list_1h",
|
|
|
+ "share_uv_list_1h", "share_pv_list_1h", "return_uv_list_1h", "p_return_uv_list_1h"
|
|
|
+
|
|
|
+ ).foreach(key => {
|
|
|
+ if (feaJson.containsKey(key)) {
|
|
|
+ val value = ExtractorUtils.bucketCnt(feaJson.getString(key).toDouble)
|
|
|
+ feaJson.put(key, value.toString)
|
|
|
+ }
|
|
|
+ })
|
|
|
+ (label, feaJson)
|
|
|
+ })
|
|
|
+ //5 libsvm 转换
|
|
|
+ val data4 = data3.map({
|
|
|
+ case (label, feaJson) =>
|
|
|
+ val feaSet = Set(
|
|
|
+ "ctx_week", "ctx_hour", "ctx_region", "ctx_city",
|
|
|
+ "machineinfo_brand", "machineinfo_model", "machineinfo_platform", "machineinfo_system",
|
|
|
+ "u_1day_exp_cnt", "u_1day_click_cnt", "u_1day_share_cnt", "u_1day_return_cnt",
|
|
|
+ "u_3day_exp_cnt", "u_3day_click_cnt", "u_3day_share_cnt", "u_3day_return_cnt",
|
|
|
+ "total_time", "play_count_total",
|
|
|
+ "i_1day_exp_cnt", "i_1day_click_cnt", "i_1day_share_cnt", "i_1day_return_cnt",
|
|
|
+ "i_3day_exp_cnt", "i_3day_click_cnt", "i_3day_share_cnt", "i_3day_return_cnt",
|
|
|
+ "u_1day_ctr", "u_1day_str", "u_1day_rov", "u_1day_ros",
|
|
|
+ "u_3day_ctr", "u_3day_str", "u_3day_rov", "u_3day_ros",
|
|
|
+ "i_1day_ctr", "i_1day_str", "i_1day_rov", "i_1day_ros",
|
|
|
+ "i_3day_ctr", "i_3day_str", "i_3day_rov", "i_3day_ros",
|
|
|
+
|
|
|
+ "i_1day_ctr_rt", "i_1day_str_rt", "i_1day_ros_rt", "i_1day_rov_rt",
|
|
|
+ "i_1h_ctr_rt", "i_1h_str_rt", "i_1h_ros_rt", "i_1h_rov_rt"
|
|
|
+ )
|
|
|
+ val feaMap = new util.HashMap[String, String]()
|
|
|
+ feaSet.foreach(r => {
|
|
|
+ if (feaJson.containsKey(r)) {
|
|
|
+ feaMap.put(r, feaJson.getString(r))
|
|
|
+ }
|
|
|
+ })
|
|
|
+ val bytesFeatureExtractor = new OfflineVlogShareLRFeatureExtractorV2()
|
|
|
+ bytesFeatureExtractor.makeFeature4String(feaMap)
|
|
|
+ val featureMap = bytesFeatureExtractor.featureMap
|
|
|
+ label + "\t" + featureMap.entries().map(r => r.getValue.getIdentifier + ":1").mkString("\t")
|
|
|
+
|
|
|
+ })
|
|
|
+
|
|
|
+ // 7 保存数据到hdfs
|
|
|
+ hdfsPath = savePath + "/" + partition
|
|
|
+ if (hdfsPath.nonEmpty && hdfsPath.startsWith("/dw/recommend/model/")){
|
|
|
+ println("删除路径并开始数据写入:" + hdfsPath)
|
|
|
+ MyHdfsUtils.delete_hdfs_path(hdfsPath)
|
|
|
+ if (ifRepart == 0){
|
|
|
+ data4.saveAsTextFile(hdfsPath, classOf[GzipCodec])
|
|
|
+ }else{
|
|
|
+ data4.repartition(ifRepart).saveAsTextFile(hdfsPath, classOf[GzipCodec])
|
|
|
+ }
|
|
|
+ }else{
|
|
|
+ println("路径不合法，无法写入:" + hdfsPath)
|
|
|
+ }
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ def func(record: Record, schema: TableSchema): Record = {
|
|
|
+ record
|
|
|
+ }
|
|
|
+
|
|
|
+}
|
zhangbo