Output from get_dummies() should default to np.int8

[johne13]

The output columns from get_dummies() are binary by nature, changing this default from float64 to int8 would save 7 bytes per dummy.

df=pd.DataFrame({ 'x':[1,2,3] })
pd.get_dummies(df.x)

[bwillers]

Working on this now, hitting up against the fact that SparseDataFrame only supports floating point data... Shall we have it return uint8s for regular frames and float64s for sparse frames?

[Lunran]

Isn't this a duplicate of the following?

• BUG: dtype compat with get_dummies BUG: dtype compat with get_dummies #8725
• BUG: int dtype for get_dummies BUG: int dtype for get_dummies #13796

[jreback]

yep looks that way.