带有多个列定义的熊猫read_table

Question

I have a code generating text data, where the diagnostic output is appended to a single text file over the course of the run. Depending on how I set it up there will be different measurements taken, with an associated header row at the start of each run. The output resembles this:

# time diagnostic_1, diagnostic_2
0.3 0.25376334 0.07494259
1.7 0.3407481 0.03018158
2.2 0.45349798 0.85539953
3.4 0.22368132 0.52276335
4.8 0.17906047 0.40659944
# time diagnostic_1, diagnostic_3
3.4 0.65968555 0.67085918
4.8 0.2122165 0.80855038
5.1 0.96943873 0.41903639
6.8 0.16242912 0.91949807
7.0 0.68513815 0.22881037
8.8 0.83304083 0.02394251
9.2 0.01699944 0.58386401
# time diagnostic_2, diagnostic_3
8 0.79595325  0.8913367 
9 0.46277533  0.47859048
10 0.30773957  0.64765873
11 0.19077614  0.39109832
12 0.0020474  0.44365015

Is there a way to have pandas.read_table return after reading a specified string rather than after a specified number of lines? The work around I have right now is to do a first pass with grep to find where the splits are, and load the arrays using numpy.loadtxt

from subprocess import check_output
import numpy as np
import pandas as pd
from itertools import cycle

fname = 'foo'
headerrows = [int(s.split(b':')[0])
              for s in check_output(['grep', '-on', '^#', fname]).split()]
# -1 to the range, because the header row is read separately
limiters = [range(a, b-1) for a, b in zip(headerrows[:-1],   headerrows[1:])]
limiters += [cycle([True, ]), ]

nameses = [['t', 'diagnostic_1', 'diagnostic_2'],
           ['t', 'diagnostic_1', 'diagnostic_3'],
           ['t', 'diagnostic_2', 'diagnostic_3']]
dat = []
with open(fname, 'r') as fobj:
    for names, limit in zip(nameses, limiters):
        line = fobj.readline()
        dat.append(pd.DataFrame(np.loadtxt((s for i, s in zip(limit, fobj))),
                                columns=names))

Answer 1

Full script that spits out a dataframe with the information I want. The monkey business with updating and dropping columns is necessary to keep the composite index. retval.merge(dset, how='outer') Give the same columns, but an integer index.

from subprocess import check_output
import numpy as np
import pandas as pd
from itertools import cycle

fname = 'foo'
headerrows = [int(s.split(b':')[0])
              for s in check_output(['grep', '-on', '^#', fname]).split()]
# subtract one because header column is read separately
limiters = [range(a, b-1) for a, b in zip(headerrows[:-1], headerrows[1:])]
limiters += [cycle([True, ]), ]

nameses = [['t', 'diagnostic_1', 'diagnostic_2'],
           ['t', 'diagnostic_1', 'diagnostic_3'],
           ['t', 'diagnostic_2', 'diagnostic_3']]

with open(fname, 'r') as fobj:
    for names, limit in zip(nameses, limiters):
        line = fobj.readline()
        dset = pd.DataFrame(np.loadtxt((line for i, line in zip(limit, fobj))),
                            columns=names)
        dset.set_index('t', inplace=True)
        # if the return value already exists, merge in the new dataset
        try:
            retval = retval.merge(dset, how='outer',
                                  left_index=True, right_index=True,
                                  suffixes=('', '_'))
            for col in (c for c in retval.columns if not c.endswith('_')):
                upd = ''.join((col, '_'))
                try:
                    retval[col].update(retval[upd])
                    retval.drop(upd, axis=1, inplace=True)
                except KeyError:
                    pass
        except NameError:
            retval = dset
print(retval)