如何在 Spark 中使用用户定义模式创建 DataFrame

Question

I wanted to create on DataFrame with a specified schema in Python. Here is the process that i have done so far.

1. I have Sample.parm file, where i have defined schema like as below: Account_type,string,True

2. I have written python script sample.py to read sample.parm file,generate the schema based on sample.parm file and then generate dataframe based on user defined schema.

d

def schema():
    with open('<path>/sample.parm','r') as parm_file:
        reader=csv.reader(parm_file,delimiter=",")
        filteredSchema = []
        for fieldName in reader:
            if fieldName[1].lower() == "decimal":
               filteredSchema.append([fieldName[0], DecimalType(),fieldName[2]])
            elif fieldName[1].lower() == "string":
               filteredSchema.append([fieldName[0], StringType(),fieldName[2]])
            elif fieldName[1].lower() == "integer":
               filteredSchema.append([fieldName[0], IntegerType(),fieldName[2]])
            elif fieldName[1].lower() == "date":
               filteredSchema.append([fieldName[0], DateType(),fieldName[2]])
            elif fieldName[1].lower() == "byte":
               filteredSchema.append([fieldName[0], ByteType(),fieldName[2]])
            elif fieldName[1].lower() == "boolean":
               filteredSchema.append([fieldName[0], BooleanType(),fieldName[2]])
            elif fieldName[1].lower() == "short":
               filteredSchema.append([fieldName[0], ShortType(),fieldName[2]])
            elif fieldName[1].lower() == "long":
               filteredSchema.append([fieldName[0], LongType(),fieldName[2]])
            elif fieldName[1].lower() == "double":
               filteredSchema.append([fieldName[0], DoubleType(),fieldName[2]])
            elif fieldName[1].lower() == "float":
               filteredSchema.append([fieldName[0], FloatType(),fieldName[2]])
            elif fieldName[1].lower() == "timestamp":
               filteredSchema.append([fieldName[0], TimestampType(),fieldName[2]])
 struct_schema = [StructField(line[0], line[1], line[2]) for line in filteredSchema]
 schema=StructTpe(struct_schema)
 return schema

def create_dataframe(path):
    val=spark.read.schema(schema()).csv(path, sep='\t')
    print(val.take(1))

but getting error like : pyspark.sql.utils.IllegalArgumentException: u'Failed to convert the JSON string \\'{"metadata":{},"name":"account_type","nullable":"True","type":"string"}\\' to a field.'

can you please anyone help me to figure it out? appreciate your help

Answer 1

I think JSON build is not correct- the metadata is empty,"type" and "field" are missing. Please try the following JSON for your schema.

{"type":"struct","fields":[{"name":"account_type","type":"string","nullable":true,"metadata":{"name":"account_type","scale":0}}]}