将键值对读入Pandas

Question

Pandas makes it really easy to read a CSV file:

pd.read_table('data.txt', sep=',')

Does Pandas having something similar for a file with key-value pairs? I came-up with this:

pd.DataFrame([dict([p.split('=') for p in l.split(',')]) for l in open('data.txt')])

If not built-in, then perhaps something more idiomatic?

The file of interest looks like this:

symbol=ESM3,exchange=GLOBEX,timestamp=1365428525690751,price=1548.00,quantity=551
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525697183,price=1548.00,quantity=551
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525714498,price=1548.00,quantity=551
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525734967,price=1548.00,quantity=551
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525735567,price=1548.00,quantity=555
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525735585,price=1548.00,quantity=556
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525736116,price=1548.00,quantity=556
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525740757,price=1548.00,quantity=556
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525748502,price=1548.00,quantity=556
symbol=ESM3,exchange=GLOBEX,timestamp=1365428525748952,price=1548.00,quantity=557

It has the exact same keys on every line, and in the same order. There are no null values. The table to be generated is:

  exchange    price quantity symbol         timestamp
0   GLOBEX  1548.00    551\n   ESM3  1365428525690751
1   GLOBEX  1548.00    551\n   ESM3  1365428525697183
2   GLOBEX  1548.00    551\n   ESM3  1365428525714498
3   GLOBEX  1548.00    551\n   ESM3  1365428525734967
4   GLOBEX  1548.00    555\n   ESM3  1365428525735567
5   GLOBEX  1548.00    556\n   ESM3  1365428525735585
6   GLOBEX  1548.00    556\n   ESM3  1365428525736116
7   GLOBEX  1548.00    556\n   ESM3  1365428525740757
8   GLOBEX  1548.00    556\n   ESM3  1365428525748502
9   GLOBEX  1548.00    557\n   ESM3  1365428525748952

(I can remove the \\n from quantity with an rstrip() after I've brought it in.)

Answer 1

If you know the key names beforehand and if the names always appear in the same order, then you could use a converter to chop off the key names, and then use the names parameter to name the columns:

import pandas as pd

def value(item):
    return item[item.find('=')+1:]

df = pd.read_table('data.txt', header=None, delimiter=',',
                   converters={i:value for i in range(5)},
                   names='symbol exchange timestamp price quantity'.split())
print(df)

on your posted data yields

  symbol exchange         timestamp    price quantity
0   ESM3   GLOBEX  1365428525690751  1548.00      551
1   ESM3   GLOBEX  1365428525697183  1548.00      551
2   ESM3   GLOBEX  1365428525714498  1548.00      551
3   ESM3   GLOBEX  1365428525734967  1548.00      551
4   ESM3   GLOBEX  1365428525735567  1548.00      555
5   ESM3   GLOBEX  1365428525735585  1548.00      556
6   ESM3   GLOBEX  1365428525736116  1548.00      556
7   ESM3   GLOBEX  1365428525740757  1548.00      556
8   ESM3   GLOBEX  1365428525748502  1548.00      556
9   ESM3   GLOBEX  1365428525748952  1548.00      557

Answer 2

I'm not sure what the best way to do this is, but assuming that the delimiters aren't found in the values -- it hurts my brain to think of the corner cases -- then something like this isn't super-elegant but is straightforward:

>>> df = pd.read_csv("esm.csv", sep=",|=", header=None)
>>> df2 = df.ix[:,1::2]
>>> df2.columns = list(df.ix[0,0::2])
>>> df2
  symbol exchange         timestamp  price  quantity
0   ESM3   GLOBEX  1365428525690751   1548       551
1   ESM3   GLOBEX  1365428525697183   1548       551
2   ESM3   GLOBEX  1365428525714498   1548       551
3   ESM3   GLOBEX  1365428525734967   1548       551
4   ESM3   GLOBEX  1365428525735567   1548       555
5   ESM3   GLOBEX  1365428525735585   1548       556
6   ESM3   GLOBEX  1365428525736116   1548       556
7   ESM3   GLOBEX  1365428525740757   1548       556
8   ESM3   GLOBEX  1365428525748502   1548       556
9   ESM3   GLOBEX  1365428525748952   1548       557

Basically, read it in, and then do the pivot yourself, keeping every other element and then fixing the column names.