Data type of pandas column changes to object when it's passed to a function via apply?

It appears to be due to an optimization in DataFrame._apply_standard. The "fast path" in the code of that method creates an output Series whose dtype is the dtype of df.values, which in your case is object since the DataFrame is of mixed type. If you pass reduce=False to your apply call, the result is correct:

>>> df.apply(lambda col: col.dtype, reduce=False)
floatcol     float64
stringcol     object
dtype: object

(I must say that it is not clear to me how this behavior of reduce jibes with the documentation.)