recommend-emr-dataprocess/ad/01_ad_model_update_everyday.sh

#!/bin/sh
set -x

# 0 全局变量/参数
originDataSavePath=/dw/recommend/model/31_ad_sample_data_v3_auto/
bucketFeatureSavePath=/dw/recommend/model/33_ad_train_data_v3_auto/
model_name=model_bkb8_v3
today="$(date +%Y%m%d)"
today_early_1="$(date -d '1 days ago' +%Y%m%d)"

LAST_MODEL_HOME=/root/zhaohp/model_online
MODEL_PATH=/root/zhaohp/recommend-emr-dataprocess/model
PREDICT_PATH=/root/zhaohp/recommend-emr-dataprocess/predict
HADOOP=/opt/apps/HADOOP-COMMON/hadoop-common-current/bin/hadoop
FM_HOME=/root/sunmingze/alphaFM
OSS_PATH=oss://art-recommend.oss-cn-hangzhou.aliyuncs.com/zhangbo/
max_hour=17
max_minute=00

export SPARK_HOME=/opt/apps/SPARK2/spark-2.4.8-hadoop3.2-1.0.8
export PATH=$SPARK_HOME/bin:$PATH
export HADOOP_CONF_DIR=/etc/taihao-apps/hadoop-conf
export JAVA_HOME=/usr/lib/jvm/java-1.8.0

start_time=$(date "+%Y-%m-%d %H:%M:%S")
elapsed=0
LOG_PREFIX=广告模型自动更新任务


# 1 判断依赖的数据表是否生产完成
source /root/anaconda3/bin/activate py37
while true; do
  python_return_code=$(python ad/ad_utils.py --excute_program check_ad_origin_hive --partition ${today} --hh 10)
  elapsed=$(($(date +%s -d "${start_time}") - $(date +%s -d "+%Y-%m-%d %H:%M:%S")))

  if [ "$python_return_code" -eq 0 ]; then
    break
  fi
  echo "Python程序返回非0值，等待五分钟后再次调用。"
  sleep 300
  current_hour=$(date +%H)
  current_minute=$(date +%M)
  if (( current_hour > max_hour || (current_hour == max_hour && current_minute >= max_minute) )); then
    msg="大数据数据生产校验失败 \n\t分区: ${today}10"
    echo -e "$LOG_PREFIX -- 大数据数据生产校验 -- ${msg}"
    /root/anaconda3/bin/python ad/ad_monitor_util.py --level error --msg "$msg" --start "$start_time" --elapsed "$elapsed"
    exit 1
  fi
done
echo "$LOG_PREFIX -- 大数据数据生产校验 -- 大数据数据生产校验通过: 耗时 $elapsed"


# 2 原始特征生成
/opt/apps/SPARK2/spark-2.4.8-hadoop3.2-1.0.8/bin/spark-class2 org.apache.spark.deploy.SparkSubmit \
--class com.aliyun.odps.spark.zhp.makedata_ad.makedata_ad_31_originData_20240620 \
--master yarn --driver-memory 1G --executor-memory 2G --executor-cores 1 --num-executors 16 \
./target/spark-examples-1.0.0-SNAPSHOT-shaded.jar \
tablePart:64 repartition:16 \
beginStr:${today_early_1}00 endStr:${today}10 \
savePath:${originDataSavePath} \
table:alg_recsys_ad_sample_all filterHours:00,01,02,03,04,05,06,07 \
idDefaultValue:0.01

elapsed=$(($(date +%s -d "${start_time}") - $(date +%s -d "+%Y-%m-%d %H:%M:%S")))
if [ $? -ne 0 ]; then
   msg="Spark原始样本生产任务执行失败"
   echo "$LOG_PREFIX -- 原始样本生产 -- $msg: 耗时 $elapsed"
   /root/anaconda3/bin/python ad/ad_monitor_util.py --level error --msg "$msg" --start "$start_time" --elapsed "$elapsed"
   exit 1
else
   echo "$LOG_PREFIX -- 原始样本生产 -- Spark原始样本生产任务执行成功: 耗时 $elapsed"
fi


# 3 特征分桶
/opt/apps/SPARK2/spark-2.4.8-hadoop3.2-1.0.8/bin/spark-class2 org.apache.spark.deploy.SparkSubmit \
--class com.aliyun.odps.spark.zhp.makedata_ad.makedata_ad_33_bucketData_20240622 \
--master yarn --driver-memory 2G --executor-memory 4G --executor-cores 1 --num-executors 16 \
./target/spark-examples-1.0.0-SNAPSHOT-shaded.jar \
beginStr:${today_early_1} endStr:${today} repartition:100 \
filterNames:adid_,targeting_conversion_ \
readPath:${originDataSavePath} \
savePath:${bucketFeatureSavePath}

elapsed=$(($(date +%s -d "${start_time}") - $(date +%s -d "+%Y-%m-%d %H:%M:%S")))
if [ $? -ne 0 ]; then
   msg="Spark特征分桶处理任务执行失败"
   echo "$LOG_PREFIX -- 特征分桶处理任务 -- $msg: 耗时 $elapsed"
   /root/anaconda3/bin/python ad/ad_monitor_util.py ${msg}
   exit 1
else
   echo "$LOG_PREFIX -- 特征分桶处理任务 -- spark特征分桶处理执行成功: 耗时 $elapsed"
fi


# 4 模型训练
$HADOOP fs -text ${bucketFeatureSavePath}/${today_early_1}/* | ${FM_HOME}/bin/fm_train -m ${MODEL_PATH}/${model_name}_${today_early_1}.txt -dim 1,1,8  -im ${LAST_MODEL_HOME}/model_online.txt -core 8
if [ $? -ne 0 ]; then
   echo "模型训练失败"
   /root/anaconda3/bin/python ad/ad_monitor_util.py "广告模型训练失败"
   exit 1
fi


# 5 对比AUC
$HADOOP fs -text ${bucketFeatureSavePath}/${today}/* | ${FM_HOME}/bin/fm_predict -m ${LAST_MODEL_HOME}/model_online.txt -dim 8 -core 8 -out ${PREDICT_PATH}/${model_name}_${today}_online.txt
$HADOOP fs -text ${bucketFeatureSavePath}/${today}/* | ${FM_HOME}/bin/fm_predict -m ${MODEL_PATH}/${model_name}_${today_early_1}.txt -dim 8 -core 8 -out ${PREDICT_PATH}/${model_name}_${today}_new.txt

# 5.1 计算线上模型的AUC
online_auc=`cat ${PREDICT_PATH}/${model_name}_${today}_online.txt | /root/sunmingze/AUC/AUC`
if [ $? -ne 0 ]; then
   echo "线上模型AUC计算失败"
   /root/anaconda3/bin/python ad/ad_monitor_util.py "线上模型AUC计算失败"
   exit 1
fi

# 5.2 计算新模型的AUC
new_auc=`cat ${PREDICT_PATH}/${model_name}_${today}_new.txt | /root/sunmingze/AUC/AUC`
if [ $? -ne 0 ]; then
   echo "新模型AUC计算失败"
   /root/anaconda3/bin/python ad/ad_monitor_util.py "新模型AUC计算失败"
   exit 1
fi

echo "AUC比对: 线上模型的AUC: ${online_auc}, 新模型的AUC: ${new_auc}"

# 5.3 计算新模型与线上模型的AUC差值
auc_diff=$(echo "$online_auc - $new_auc" | bc -l)

# 5.4 获取差值的绝对值
auc_diff_abs=$(echo "sqrt(($auc_diff)^2)" | bc -l)

# 5.5 如果差值的绝对值小于0.005且新模型的AUC大于0.73, 则更新模型
if (( $(echo "${online_auc} <= ${new_auc}" | bc -l) )); then
    echo "新模型优于线上模型: 线上模型AUC: ${online_auc}, 新模型AUC: ${new_auc}"
    /root/anaconda3/bin/python ad/ad_monitor_util.py "新模型优于线上模型: 线上模型AUC: ${online_auc}, 新模型AUC: ${new_auc}"
elif (( $(echo "$auc_diff_abs < 0.005" | bc -l) )) && (( $(echo "$new_auc >= 0.73" | bc -l) )); then
    echo "新模型与线上模型差值小于阈值0.005: 线上模型AUC: ${online_auc}, 新模型AUC: ${new_auc}, 差值为: $auc_diff_abs"
    /root/anaconda3/bin/python ad/ad_monitor_util.py "新模型与线上模型差值小于阈值0.005: 线上模型AUC: ${online_auc}, 新模型AUC: ${new_auc}, 差值为: $auc_diff_abs"
else
    echo "新模型与线上模型差值大于等于阈值0.005: 线上模型AUC: ${online_auc}, 新模型AUC: ${new_auc}, 差值为: $auc_diff_abs"
    /root/anaconda3/bin/python ad/ad_monitor_util.py "新模型与线上模型差值大于等于阈值0.005: 线上模型AUC: ${online_auc}, 新模型AUC: ${new_auc}, 差值为: $auc_diff_abs"
    exit 1
fi


# 6 模型格式转换
cat ${MODEL_PATH}/${model_name}_${today_early_1}.txt |
awk -F " " '{
    if (NR == 1) {
        print $1"\t"$2
    } else {
        split($0, fields, " ");
        OFS="\t";
        line=""
        for (i = 1; i <= 10 && i <= length(fields); i++) {
            line = (line ? line "\t" : "") fields[i];
        }
        print line
    }
}' > ${MODEL_PATH}/${model_name}_${today_early_1}_change.txt
if [ $? -ne 0 ]; then
   echo "新模型文件格式转换失败"
   /root/anaconda3/bin/python ad/ad_monitor_util.py "新模型文件格式转换失败"
   exit 1
fi

# 7 模型文件上传OSS
online_model_path=${OSS_PATH}/${model_name}.txt
$HADOOP fs -test -e ${online_model_path}
if [ $? -eq 0 ]; then
    echo "数据存在, 先删除。"
    $HADOOP fs -rm -r -skipTrash ${online_model_path}
else
    echo "数据不存在"
fi

$HADOOP fs -put ${MODEL_PATH}/${model_name}_${today_early_1}_change.txt ${online_model_path}
if [ $? -eq 0 ]; then
   echo "推荐模型文件至OSS成功"
else
   echo "推荐模型文件至OSS失败"
   /root/anaconda3/bin/python ad/ad_monitor_util.py "推荐模型文件至OSS失败"
   exit 1
fi

# 7.3 本地保存最新的线上使用的模型，用于下一次的AUC验证
cp -f ${LAST_MODEL_HOME}/model_online.txt ${LAST_MODEL_HOME}/model_online_$(date +\%Y\%m\%d).txt
cp -f ${MODEL_PATH}/${model_name}_${today_early_1}.txt ${LAST_MODEL_HOME}/model_online.txt
if [ $? -ne 0 ]; then
   echo "模型备份失败"
   /root/anaconda3/bin/python ad/ad_monitor_util.py "模型备份失败 - 最新模型地址: ${MODEL_PATH}/${model_name}_${today_early_1}.txt"
   exit 1
fi

# 32 16 * * * cd /root/zhangbo/recommend-emr-dataprocess && /bin/sh ./ad/01_ad_model_update_everyday.sh > logs/01_update_eventday$(date +\%Y-\%m-\%d_\%H).log 2>&1