pyspark dataframe列的合并与拆分实例

from pyspark.sql import SparkSession
spark = SparkSession.builder \
    .master("local") \
    .appName("dataframe_split") \
    .config("spark.some.config.option", "some-value") \
    .getOrCreate()
 
sc = spark.sparkContext
df = spark.read.csv('hdfs://master:9000/dataset/dataframe_split.csv', inferSchema=True, header=True)
df.show(3)

from pyspark.sql.functions import split, explode, concat, concat_ws
df_split = df.withColumn("s", split(df['score'], " "))
df_split.show()

first_row = df.first()
numAttrs = len(first_row['score'].split(" "))
print("新增列的个数", numAttrs)
attrs = sc.parallelize(["score_" + str(i) for i in range(numAttrs)]).zipWithIndex().collect()
print("列名：", attrs)
for name, index in attrs:
    df_split = df_split.withColumn(name, df_split['s'].getItem(index))
df_split.show()

df_explode = df.withColumn("e", explode(split(df['score'], " ")))
df_explode.show()

df_concat = df_split.withColumn("score_concat", concat(df_split['score_0'], \
                                                       df_split['score_1'], df_split['score_2'], df_split['score_3']))
df_concat.show()

df_ws = df_split.withColumn("score_concat", concat_ws('-', df_split['score_0'], \
                                                       df_split['score_1'], df_split['score_2'], df_split['score_3']))
df_ws.show()

#DataFrame 数据格式:每个用户对每部电影的评分 userID 用户ID,movieID 电影ID,rating评分
df=spark.sparkContext.parallelize([[15,399,2], \
                                   [15,1401,5], \
                                   [15,1608,4], \
                                   [15,20,4], \
                                   [18,100,3], \
                                   [18,1401,3], \
                                   [18,399,1]])\
                    .toDF(["userID","movieID","rating"])
#pivot 多行转多列
resultDF = df.groupBy("userID").pivot("movieID").sum("rating").na.fill(-1)
#结果
resultDF.show()

from pyspark.sql.functions import length, col, lit, size
df.withColumn("length_col", length(col("existing_str_col"))) # 将existing_str_col的长度生成新列
df.withColumn("constant_col", lit("hello")) # 生成一列常量
df.withColumn("size_col", size(col("existing_array_col"))) # 将existing_array_col的元素个数生成新列

from pyspark.sql.functions import col
df = df.select(col("col_1").cast("string"), col("col_2").alias("col_2_")) # 选择col_1列和col_2列，并将col_1列转换为string格式，将col_2列重命名为col_2_,此时不再存在col_2

from pyspark.sql.functions import concat_ws
 
df = df.withColumn("concat_col", concat_ws("_", df.col_1, df.col_2)) # 使用_将col_1和col_2连接起来，构成新列concat_col

from pyspark.sql.functions import split
 
df = df.withColumn("split_col", split(df.col, "-")) #按照-将df中的col列分割，此时split_col时一个list，后续或者配合filter(length(...))使用

from pyspark.sql.functions import mean
 
col_mean = df.select(mean(col)).collect()[0][0]

print(df.take(5)) #交互式的pyspark shell中，等价于df.show(5)

print(df.count()) #统计行数量

from pyspark.sql.functions import col
df = df.filter(col("col_1")==col("col_2")) #保留col_1等于col_2的行

df.na.drop("all") # 只有当所有列都为空时，删除该行
df.na.drop("any") # 任意列为空时，删除该行
df.na.drop("all", colsubset=["col_1","col_2"]) # 当col_1和col_2都为空时，删除该行

df = df.distinct() # 删除所有列值相同的重复行
df = df.dropDuplicates(["date", "count"]) # 删除date, count两列值相同的行

from pyspark.sql.functions import explode, split
 
df = df.withColumn("sub_str", explode(split(df["str_col"], "_"))) # 将str_col按-拆分成list，list中的每一个元素成为sub_str,与原行中的其他列一起组成新的行

df.na.fill({"col_name":fill_content}) # 用fill_content填补col_name列的空值

from pyspark.sql.functions import monotonically_increasing_id
 
df = df.withColumn("id", monotonically_increasing_id())

df_3 = df_1.join(df_2, df_1.col_1==df_2.col_2) # inner join, 只有当df_1中的col_1列值等于df_2中的col_2时，才会拼接
df_4 = df_1.join(df_2, df_1.col_1==df_2.col_2, "left") # left join, 当df_1中的col_1值不存在于df_2中时，仍会拼接，凭借值填充null

df3 = df1.union(df2)

groupBy
from pyspark.sql.functions import concat_ws, split, explode, collect_list, struct
 
concat_df = concat_df.groupBy("sample_id", "sample_date").agg(collect_list('feature').alias("feature_list")) # 将同sample_id, sample_date的行聚合成组，feature字段拼成一个list，构成新列feature_list。agg配合groupBy使用，效果等于select。此时concat_df只有两列：sample_id和feature_list。
concat_tuple_df = concat_df.groupBy("sample_id", "sample_date").agg(collect_list(struct("feature", "owner")).alias("tuple")) # 将同sample_id, sample_date的行聚合成组， (feature, owner)两个字段拼成一个单位，组内所有单位拼成一个list，构成新列tuple

from pyspark.sql.window import Window
from pyspark.sql.functions import col, row_number
 
windowSpec = Window.partitionBy(df.id, df.date).orderBy(col("price").desc(), col("discount").asc()) # 相同id,date的行被聚成组，组内按照price降序，discount升序进行排列
df = df.withColumn("rank", row_number().over(windowSpec)) #为排序之后的组进行组内编号
df = df.filter(df.rank<=1) # 取组内top-1行

from pyspark.sql import SparkSession
from pyspark import SparkContext
 
sc = SparkContext(appName="test_rw")
sc_session = SparkSession(sc)
df.write.mode("overwrite").options(header="true").csv(output_path)
df = sc_session.csv.read(input_path, header=True)

from pyspark import SparkContext
from pyspark.sql import SparkSession
 
sc = SparkContext(appName='get_sample')
sc_session = SparkSession(sc)
 
sample_df.createOrReplaceTempView("item_sample_df")
sample_df = sc_session.sql(
        '''
            select sample_id
                ,label
                ,type_ as type
                ,split(item_id, "_")[2] as owner
                ,ftime
            from item_sample_df
        ''')

from pysprak.sql.functions import udf, col
from pyspark.sql.types import StringType, ArrayType, StructField, StructType
 
 
def simple_func(v1, v2):
    pass
    # return str
 
simple_udf = udf(my_func, StringType())
 
df = df.withColumn("new", simple_udf(df["col_1"], df["col_2"]))
 
 
 
# 复杂type
 
def get_entity_func():
    pass
    # return str_list_1, str_list_2
 
entity_schema = StructType([
                    StructField("location", ArrayType(StringType()), True),
                    StructField("nondigit", ArrayType(StringType()), True)
                ])
 
get_entity_udf = udf(get_entity_func, entity_schema)

from pyspark import SparkContext
from pyspark.sql import SparkSession
from pyspark.sql.types import StructType, StructField, StringType, FloatType
 
 
sc = SparkContext(appName="rdd2df")
sc_session = SparkSession(sc)
 
rdd = df.rdd # df转rdd, 注意每列仍带header，要map(lambda line: [line.id, line.price])才可以转换成不带header
 
schema = StructType([
                    StructField("id", StringType(), True),
                    StructField("price", FloatType(), True)
                    ])
df = sc_session.createDataFrame(rdd, schema) # rdd转df