我有一个数据帧，我一度转换为rdd来执行自定义计算 . 在使用UDF（创建新列）完成此操作之前，我注意到这很慢 . 因此我转换为RDD并再次返回，但是我注意到在将rdd转换为dataframe期间执行似乎停滞不前 .

conf = SparkConf().setMaster(server_location).setAppName("MoleculesTests")

ss = SparkSession.builder.config(conf = conf).getOrCreate()
ss.sparkContext.addPyFile("helpers.py")

def mapper(line):
    fields = line.split(' ')
    return Row(name=str(fields[0]))

def calculate_tanimoto(smiles1,smiles2):
    try:
        mol1 = Chem.MolFromSmiles(smiles1)
        mol2 = Chem.MolFromSmiles(smiles2)
        fp1 = AllChem.GetMorganFingerprintAsBitVect(mol1, 2)
        fp2 = AllChem.GetMorganFingerprintAsBitVect(mol2, 2)
        similarity = DataStructs.FingerprintSimilarity(fp1,fp2, metric=DataStructs.TanimotoSimilarity)

        return similarity
    except Exception as e:
        print (str(e))
        print ("Error Smiles1", smiles1, " 2", smiles2)


CREATE_VECTORS = False
SIMILARITY_THRESHOLD = 0.3
dataFile = '../mols/compounds18.smi'
lines = ss.sparkContext.textFile(dataFile)

smiles = lines.map(mapper)
schemaSmiles = ss.createDataFrame(smiles).cache()
schemaSmiles.createOrReplaceTempView("smiles")

#some basic filtering
valid_smiles = ss.sql("SELECT * FROM smiles WHERE name != 'smiles'")
valid_smiles_id = valid_smiles.select("*").withColumn("id", monotonically_increasing_id())

#cartesian join and then filtering to get only upper triangle of a similarity matrix, with result being Source_Id, Source_description, Target_id, Target_description
combinations = valid_smiles_id.alias("source").join(valid_smiles_id.alias("target") )\
    .where("source.Id <= target.Id")\
    .select(f.col("source.Id").alias("source_id"), f.col("source.Name").alias("source_smile"), f.col("target.Id").alias("target_id"),f.col("target.Name").alias("target_smile"))


 #Change to rdd to perform calculate_tanimoto using source and target descriptions 
combinations_rdd = combinations.rdd.map(tuple)
similarities_fp = combinations_rdd.map(lambda (source_id, source_smiles,target_id,target_smiles): (source_id, target_id, calculate_tanimoto(source_smiles, target_smiles)))\
                              .filter(lambda (a,b,c): c >= SIMILARITY_THRESHOLD).cache()


schema = StructType([StructField("source",IntegerType(), False),StructField("target",IntegerType(), False),StructField("tanimoto",StringType(), False)])
#change back to Dataframe, execution seems to get stuck here
combinations_sim = ss.createDataFrame(similarities_fp,schema=schema).cache()
print(combinations_sim.show(n=10))
combinations_sim = combinations_sim.groupby(combinations_sim.source_id).agg(f.collect_set("target_id"))

为2500种化合物运行此类似乎卡住了 .

apache-spark pyspark