我有一个具有以下结构的.csv:
date_begin,date_end,name,name_code,active_accounts,transaction_amount,transaction_count
1/1/2008,1/31/2008,Name_1,1001,"123,456","$7,890,123.45","67,890"
2/1/2008,2/29/2008,Name_1,1001,"43,210","$987,654.32","109,876"
3/1/2008,3/31/2008,Name_1,1001,"485,079","$1,265,789,433.98","777,888"
...
12/1/2008,12/31/2008,Name_1,1001,"87,543","$432,098,987","87,987"
1/1/2008,1/31/2008,Name_2,1002,"268,456","$890,123.45","97,890"
2/1/2008,2/29/2008,Name_2,1002,"53,210","$987,654.32","109,876"
...
etc
我正在尝试使用以下代码将它们读入pandas数据框:
import pandas as pd
data = pd.read_csv('my_awesome_csv.csv'),parse_dates=[[0,1]],
infer_datetime_format=True)
这很好用,除了我想控制每列中的数据类型。当我在解释器中运行以下代码时,我发现引号中的数字不会被识别为数字,无论是美元还是其他。
In [10]: data.dtypes
Out[10]:
date_begin_date_end object
name object
name_code int64
active_accounts object # Problem, I want this to be a number
transaction_amount object # Ditto, I want this to be a number (it's a dollar amount)
transaction_count object # Still a number!
dtype: object
我已经在Pandas csv documentation中做了一些窥探,但是在csv中用逗号和美元符号保存为字符串时,我没有找到关于声明类型的内容。我的最终目标是能够对这些列中的值进行一些算术运算。
有什么想法吗?
答案 0 :(得分:10)
您可以在调用后使用vectorized string methods来解析这些列
read_csv
:
import pandas as pd
import decimal
D = decimal.Decimal
data = pd.read_csv('data', parse_dates=[[0,1]], infer_datetime_format=True)
for col in ('active_accounts', 'transaction_count'):
data[col] = data[col].str.replace(r',', '').astype(int)
data['transaction_amount'] = (data['transaction_amount']
.str.replace(r'[^-+\d.]', '').astype(D))
print(data.dtypes)
# date_begin_date_end object
# name object
# name_code int64
# active_accounts int64
# transaction_amount object
# transaction_count int64
# dtype: object
print(data)
产量
date_begin_date_end name name_code active_accounts \
0 1/1/2008 1/31/2008 Name_1 1001 123456
1 2/1/2008 2/29/2008 Name_1 1001 43210
2 3/1/2008 3/31/2008 Name_1 1001 485079
3 12/1/2008 12/31/2008 Name_1 1001 87543
4 1/1/2008 1/31/2008 Name_2 1002 268456
5 2/1/2008 2/29/2008 Name_2 1002 53210
transaction_amount transaction_count
0 7890123.45 67890
1 987654.32 109876
2 1265789433.98 777888
3 432098987 87987
4 890123.45 97890
5 987654.32 109876
PS。 read_csv
确实有converters
parameter
您可以使用它来提供解析有问题列的函数。这些
每个字符串调用一次函数。如果你有很多行,那可能需要很多Python函数
调用。使用矢量化字符串方法处理列,如上所示
应该快得多。
import pandas as pd
import re
import decimal
D = decimal.Decimal
def make_parser(cls):
def parse_commas(text):
return cls(re.sub(r'[^-+\d.]', '', text))
return parse_commas
to_int = make_parser(int)
to_decimal = make_parser(D)
data = pd.read_csv('data', parse_dates=[[0,1]], infer_datetime_format=True
, converters={4: to_int, 5: to_decimal, 6: to_int})
print(data)
产量
date_begin_date_end name name_code active_accounts \
0 1/1/2008 1/31/2008 Name_1 1001 123456
1 2/1/2008 2/29/2008 Name_1 1001 43210
2 3/1/2008 3/31/2008 Name_1 1001 485079
3 12/1/2008 12/31/2008 Name_1 1001 87543
4 1/1/2008 1/31/2008 Name_2 1002 268456
5 2/1/2008 2/29/2008 Name_2 1002 53210
transaction_amount transaction_count
0 7890123.45 67890
1 987654.32 109876
2 1265789433.98 777888
3 432098987 87987
4 890123.45 97890
5 987654.32 109876
并且transaction_amount
列中的值为decimal.Decimals:
In [64]: data.loc[0, 'transaction_amount']
Out[64]: Decimal('7890123.45')