Closed
Description
Just putting this up there if anyone has any ideas. The filling scales with the length of the axis so this is unavoidable (well it could be done in cython but would need a 3d version of certain functions).
can think about changing the default to fill_method=None from fill_method='pad'
In [8]: index = pd.date_range(start="2000", freq="D", periods=1000)
In [9]: panel = pd.Panel(np.random.randn(100, len(index), 1000))
In [10]: %timeit panel.pct_change(1, axis="major", fill_method='pad')
1 loops, best of 3: 3.62 s per loop
In [11]: %timeit panel.pct_change(1, axis="major", fill_method=None)
1 loops, best of 3: 1.46 s per loop