Question

我正在尝试从fixedwidth文件创建一个datafrmae并加载到postgresql数据库中。我的输入文件非常庞大（~16GB）和20万条记录。因此，如果我创建数据帧，它会消耗大部分可用的RAM。这需要很长时间才能完成。所以我想到使用chunksize（使用python生成器）选项并将记录提交到表。但它失败了'AttributeError: 'generator' object has no attribute 'to_sql'错误。

受到这个答案的启发https://stackoverflow.com/a/47257676/2799214

输入文件：test_file.txt

XOXOXOXOXOXO9
AOAOAOAOAOAO8
BOBOBOBOBOBO7
COCOCOCOCOCO6
DODODODODODO5
EOEOEOEOEOEO4
FOFOFOFOFOFO3
GOGOGOGOGOGO2
HOHOHOHOHOHO1

sample.py

import pandas.io.sql as psql
import pandas as pd
from sqlalchemy import create_engine

def chunck_generator(filename, header=False,chunk_size = 10 ** 5):
    for chunk in pd.read_fwf(filename, colspecs=[[0,12],[12,13]],index_col=False,header=None, iterator=True, chunksize=chunk_size):
        yield (chunk)

def _generator( engine, filename, header=False,chunk_size = 10 ** 5):
    chunk = chunck_generator(filename, header=False,chunk_size = 10 ** 5)
    chunk.to_sql('sample_table', engine, if_exists='replace', schema='sample_schema', index=False)
    yield row

if __name__ == "__main__":
    filename = r'test_file.txt'
    engine = create_engine('postgresql://ABCD:ABCD@ip:port/database')
    c = engine.connect()
    conn = c.connection
    generator = _generator(engine=engine, filename=filename)
    while True:
       print(next(generator))
    conn.close()

错误：

    chunk.to_sql('sample_table', engine, if_exists='replace', schema='sample_schema', index=False)
AttributeError: 'generator' object has no attribute 'to_sql'

我的主要目标是提高效果。请帮我解决问题或请建议更好的方法。提前谢谢。

Answer 1

'chunck_generator'将返回一个'generator'对象而不是chunk的实际元素。您需要迭代对象以从中获取块。

>>> def my_generator(x):
...     for y in range(x):
...         yield y
...
>>> g = my_generator(10)
>>> print g.__class__
<type 'generator'>
>>> ele = next(g, None)
>>> print ele
0
>>> ele = next(g, None)
>>> print ele
1

因此，要修复代码，您只需要循环生成器

for chunk in chunck_generator(filename, header=False,chunk_size = 10 ** 5):
    yeild chunk.to_sql()

但似乎令人沮丧。我只是这样做：

import pandas.io.sql as psql
import pandas as pd
from sqlalchemy import create_engine

def sql_generator(engine, filename, header=False,chunk_size = 10 ** 5):
    frame = pd.read_fwf(
        filename, 
        colspecs=[[0,12],[12,13]],
        index_col=False,
        header=None, 
        iterator=True, 
        chunksize=chunk_size
    ):

    for chunk in frame:
        yield chunk.to_sql(
            'sample_table', 
            engine, 
            if_exists='replace', 
            schema='sample_schema', 
            index=False
        )


if __name__ == "__main__":
    filename = r'test_file.txt'
    engine = create_engine('postgresql://USEE:PWD@IP:PORT/DB')
    for sql in sql_generator(engine, filename):
        print sql

Answer 2

结论： to_sql方法加载大文件效率不高。所以我在包psycopg2中使用了copy_from方法，并在创建数据帧时使用了chunksize选项。载入980万条记录（~17GB），每条30分钟内有98列。

我删除了原始文件的原始引用（我在原帖中使用了示例文件）。

import pandas as pd
import psycopg2
import io

def sql_generator(cur,con, filename, boundries, col_names, header=False,chunk_size = 2000000):
    frame = pd.read_fwf(filename,colspecs=boundries,index_col=False,header=None,iterator=True,chunksize=chunk_size,names=col_names)
    for chunk in frame:
        output = io.StringIO()
        chunk.to_csv(output, sep='|', quoting=3, escapechar='\\' , index=False, header=False,encoding='utf-8')
        output.seek(0)
        cur.copy_from(output, 'sample_schema.sample_table', null="",sep="|")
        yield con.commit()

if __name__ == "__main__":
    boundries = [[0,12],[12,13]]
    col_names = ['col1','col2']
    filename = r'test_file.txt'  #Refer to sample file in the original post
    con = psycopg2.connect(database='database',user='username', password='pwd', host='ip', port='port')
    cur = con.cursor()
    for sql in sql_generator(cur,con, filename, boundries, col_names):
        print(sql)
    con.close()

Answer 3

我建议你这样的事情：

def _generator( engine, filename, ...):
    for chunk in pd.read_fwf(filename, ...):
        yield chunk.to_sql('sample_table', engine, ...)  # not sure about this since row was not define

for row in _generator(engine=engine, filename=filename)
    print(row)

AttributeError：'generator'对象没有属性'to_sql'使用生成器

3 个答案: