PERF: frame_repr_wide benchmark

[jreback]

Is this correct? (on current master)

In [1]: df = pandas.DataFrame(np.random.randn(10,10000))

In [2]: %timeit repr(df)
1 loops, best of 3: 9.85 s per loop

n [4]: pandas.__version__
Out[4]: '0.12.0.dev-a62ea81'

%prun repr(df)

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
  1920528    1.310    0.000    1.710    0.000 re.py:226(_compile)
  1119307    1.253    0.000    1.724    0.000 StringIO.py:208(write)
  1920528    1.092    0.000    1.092    0.000 {method 'search' of '_sre.SRE_Pattern' objects}
  1920528    1.069    0.000    3.870    0.000 re.py:139(search)
   210010    1.026    0.000    1.763    0.000 format.py:1537(just)
    80022    0.884    0.000    5.224    0.000 config.py:455(_select_options)
    60003    0.782    0.000    0.782    0.000 {method 'reduce' of 'numpy.ufunc' objects}
  1940531    0.403    0.000    0.403    0.000 {method 'get' of 'dict' objects}
    80022    0.402    0.000    0.402    0.000 {sorted}
   530020    0.345    0.000    0.345    0.000 {_codecs.utf_8_decode}

[TomAugspurger]

In [2]: df = pandas.DataFrame(np.random.randn(10,10000))

In [3]: %timeit repr(df)
1000 loops, best of 3: 1.24 ms per loop

In [4]: pandas.__version__
Out[4]: '0.12.0.dev-2c6f4c6'

And then I updated and got

In [2]: df = pandas.DataFrame(np.random.randn(10,10000))

In [3]: %timeit repr(df)
1 loops, best of 3: 13.4 s per loop

In [4]: pandas.__version__
Out[4]: '0.12.0.dev-84fb0e0'

So somewhere between those two commits.

[cpcloud]

b036a7e is the culprit, darn :/

argh. It just won't stay down. Thanks for catching this jeff.

Never in my life have I spent so much time coding myself right back to
where I started. Excepting (perhaps) Logo in primary school. God bless papert.