PERF: directly astype with numpy if series is already nansafe

[jreback]

from SO

so the null check is pretty cheap. if no nulls, then can just bypass nansafe an use the underlying numpy routine. should be a nice speedup.

``
In [13]: arr = np.random.randint(1,10,size=1000000)

In [14]: s = Series(arr)

In [15]: s.notnull().all()
Out[15]: True

In [16]: %timeit s.notnull().all()
1000 loops, best of 3: 1.35 ms per loop

In [17]: %timeit s.astype(str)
1 loops, best of 3: 2.52 s per loop

In [18]: %timeit s.values.astype(str)
10 loops, best of 3: 37.7 ms per loop

[vikram]

The time is actually not in checking for nulls.
But in ensuring that every element returned is a string.

If you did s.values.astype(str) what you get back is an object holding int. This is numpy doing the conversion, where as pandas iterates over each item and calls str(item) on it.
So if you do s.astype(str) you be an object holding str.

https://github.com/pydata/pandas/blob/master/pandas/lib.pyx#L866

So I don't think it can be fixed if we still want to returns object holding str.

Potentially https://github.com/pydata/pandas/blob/master/pandas/lib.pyx#L843
can be improved. If the array doesn't have nulls and we don't have the is_datelike
then instead of iterating, we can just return arr.astype(new_dtype)

I can sort out a pull request if there is interest.

[jbrockmendel]

s.values.astype(str) is now slightly slower than s.astype(str) (331ms vs 309ms locally). Closing.