dim_asin_title_info.py

import os
import sys

os.environ["PYARROW_IGNORE_TIMEZONE"] = "1"
sys.path.append(os.path.dirname(sys.path[0]))  # 上级目录
from utils.templates import Templates
# from ..utils.templates import Templates
from pyspark.sql.types import StringType
# 分组排序的udf窗口函数
from pyspark.sql.window import Window
from pyspark.sql import functions as F


class DimAsinTitleInfo(Templates):

    def __init__(self, site_name='us'):
        super().__init__()
        self.site_name = site_name
        self.db_save_vertical = f'dim_asin_title_info_vertical'
        self.db_save_wide = f'dim_asin_title_info_wide'
        self.spark = self.create_spark_object(app_name=f"{self.db_save_vertical}: {self.site_name}, {self.date_type}, {self.date_info}")
        self.df_theme = self.spark.sql(f"select 1+1;")
        self.df_asin_title = self.spark.sql(f"select 1+1;")
        self.df_save_vertical = self.spark.sql(f"select 1+1;")  # 竖表
        self.df_save_wide = self.spark.sql(f"select 1+1;")  # 宽表
        # 注册自定义函数 (UDF)
        self.u_theme_pattern = F.udf(self.udf_theme_pattern, StringType())
        # 其他变量
        # self.pattern = str()  # 正则匹配
        self.theme_list_str = str()  # 正则匹配
        # 分区参数
        self.partitions_by = ['site_name']
        self.partitions_num = 100

    @staticmethod
    def udf_theme_pattern(title, theme_list_str):
        found_themes = [theme.strip() for theme in eval(theme_list_str) if theme in title]
        if found_themes:
            return ','.join(set(found_themes))
        else:
            return None

    def read_data(self):
        sql = f"select id as theme_id, theme_type_en, theme_en, theme_en_lower, theme_ch from ods_theme where site_name='{self.site_name}'"
        print("sql:", sql)
        self.df_theme = self.spark.sql(sql).cache()
        self.df_theme.show(10, truncate=False)
        # sql = f"-- select asin, title as asin_volume, date_info from ods_asin_detail where site_name='{self.site_name}' and date_type='week'"  # and date_info>='2023-15'
        sql = f"select asin, title as asin_title, date_info, site_name from ods_asin_detail where site_name='{self.site_name}' and date_type='week' "  # and date_info>='2023-25' limit 10000
        print("sql:", sql)
        self.df_asin_title = self.spark.sql(sqlQuery=sql).cache()
        self.df_asin_title.show(10, truncate=False)

    def handle_data(self):
        self.handle_filter_dirty_data()
        self.handle_theme()

    def handle_filter_dirty_data(self):
        """
        过滤脏数据+保留最新的title
        """
        # 小写
        self.df_asin_title = self.df_asin_title.withColumn("asin_title_lower", F.lower(self.df_asin_title["asin_title"]))
        # 过滤空值
        self.df_asin_title = self.df_asin_title.filter("asin_title_lower is not null")
        # 过滤null和none字符串
        self.df_asin_title = self.df_asin_title.filter("asin_title_lower not in ('none', 'null', 'nan')")
        # 取最新的date_info对应的title
        window = Window.partitionBy('asin').orderBy(F.desc('date_info'))  # 按照 date_info 列进行分区，并按照 date 列进行排序
        self.df_asin_title = self.df_asin_title.withColumn('row_number', F.row_number().over(window))  # 使用窗口函数为每个分区的行编号
        self.df_asin_title = self.df_asin_title.filter(self.df_asin_title.row_number == 1).drop('row_number')  # 只保留每个分区中 row_number 最大的行，并删除 row_number 列

    def handle_theme(self):
        pdf_theme = self.df_theme.toPandas()
        theme_list = list(set(pdf_theme.theme_en_lower))
        self.theme_list_str = str([f" {theme} " for theme in theme_list])
        print("self.theme_list_str:", self.theme_list_str)
        # 匹配宽表时用到
        df_asin_title = self.df_asin_title.cache()  # 后面用作匹配asin_title

        self.df_asin_title = self.df_asin_title.withColumn("asin_title_lower", F.concat(F.lit(" "), "asin_title_lower", F.lit(" ")))  # 标题两头加空字符串用来匹配整个词
        self.df_asin_title = self.df_asin_title.withColumn("theme_en_lower", self.u_theme_pattern('asin_title_lower', F.lit(self.theme_list_str)))
        # 将列拆分为数组多列
        self.df_asin_title = self.df_asin_title.withColumn("theme_en_lower", F.split(self.df_asin_title["theme_en_lower"], ","))
        # 将数组合并到多行
        self.df_asin_title = self.df_asin_title.withColumn("theme_en_lower", F.explode(self.df_asin_title["theme_en_lower"]))
        self.df_asin_title = self.df_asin_title.join(
            self.df_theme, on=['theme_en_lower'], how='left'  # 改成inner, 这样避免正则匹配结果不准
        )
        # 1. 竖表
        self.df_save_vertical = self.df_asin_title.cache()
        print(self.df_save_vertical.columns)
        self.df_save_vertical.show(30, truncate=False)
        # self.df_save_vertical.filter("theme_en_lower is not null").show(30, truncate=False)

        # 2. 宽表
        self.df_asin_title = self.df_asin_title.drop_duplicates(['asin', 'theme_type_en', 'theme_ch'])
        self.df_asin_title = self.df_asin_title.withColumn("theme_type_en_counts", F.concat("theme_type_en", F.lit("_counts")))
        self.df_asin_title = self.df_asin_title.withColumn("theme_type_en_ids", F.concat("theme_type_en", F.lit("_ids")))
        # self.df_asin_title.filter('theme_type_en_counts is null').show(20, truncate=False)   # 没有记录
        self.df_asin_title = self.df_asin_title.filter('theme_type_en_counts is not null')
        pivot_df1 = self.df_asin_title.groupBy("asin").pivot("theme_type_en_counts").agg(
            F.expr("IFNULL(count(*), 0) AS value"))
        pivot_df1 = pivot_df1.na.fill(0)
        pivot_df2 = self.df_asin_title.groupBy("asin").pivot("theme_type_en_ids").agg(
            F.concat_ws(",", F.collect_list("theme_id")))
        pivot_df1.show(30, truncate=False)
        pivot_df2.show(30, truncate=False)
        self.df_save_wide = df_asin_title.join(
            pivot_df1, on='asin', how='left'
        ).join(
            pivot_df2, on='asin', how='left'
        )
        # self.df_save_wide.show(30, truncate=False)
        print(self.df_save_wide.columns)

    def save_data(self):
        self.reset_partitions(partitions_num=100)
        self.save_data_common(
            df_save=self.df_save_vertical,
            db_save=self.db_save_vertical,
            partitions_num=self.partitions_num,
            partitions_by=self.partitions_by
        )
        self.reset_partitions(partitions_num=100)
        self.save_data_common(
            df_save=self.df_save_wide,
            db_save=self.db_save_wide,
            partitions_num=self.partitions_num,
            partitions_by=self.partitions_by
        )


if __name__ == '__main__':
    site_name = sys.argv[1]  # 参数1：站点
    handle_obj = DimAsinTitleInfo(site_name=site_name)
    handle_obj.run()