Billion, the statement
Spark numpy and pandas can run a program, as long as you installed
First, why you want to change the dataframe Spark program with the pandas.dataframe
The former can only stand-alone operation, the latter can run clusters
Second, comparison
Jump directly to this post "Spark and Pandas in DataFrame contrast" and write well
Third, the transformation
spark —> pandas | pandas —> spark |
---|---|
pandas_df = spark_df.toPandas() | spark_df = spark.createDataFrame(pandas_df) |
Because pandas are stand-alone version of the way, that toPandas () is a stand-alone version of the way, into a distributed version:
import pandas as pd
def _map_to_pandas(rdds):
return [pd.DataFrame(list(rdds))]
def topas(df, n_partitions=None):
if n_partitions is not None: df = df.repartition(n_partitions)
df_pand = df.rdd.mapPartitions(_map_to_pandas).collect()
df_pand = pd.concat(df_pand)
df_pand.columns = df.columns
return df_pand
pandas_df = topas(spark_df)
Reference Bowen:
"spark with pandas data conversion"
"pandas and spark of dataframe Huzhuan"
Four, SparkContext in Spark2.x are integrated into SparkSession, the entire Spark podium
Reference Bowen:
"the Spark core articles -SparkContext"
"the Spark 2.0 Series SparkSession explain"