PERF: large subset of block cast

[chris-b1]

from SO

I'm guessing there's a lot of inference / checks along this path, but this is slower than I'd expect.

In [66]: df = pd.DataFrame(np.random.randint(0,1000, size=(10000, 1026)))
    ...: f = list(range(1024))
    ...: 

In [67]: %time df[f] = df[f].astype(float)
Wall time: 10.4 s

[jreback]

this is slow because this is done column by-column (that's ok), but then each column is deleted ,then added, prob with many consolidation steps in-between (each of which copy).

So a multi-assignment can be a separate case, rather than iteration and assignment; though maybe only if the rhs is in the same block (which is the case here)

[mroeschke]

Looks like this performance is a lot better now so closing

In [5]: df = pd.DataFrame(np.random.randint(0,1000, size=(10000, 1026)))

In [6]: f = list(range(1024))

In [7]: %timeit df[f] = df[f].astype(float)
251 ms ± 7.92 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)