当 dtype 为 float16 时，为什么 Pandas 不舍入？

Question

Why does Pandas not round DataFrames when the dypes are np.float16 ?

pd.DataFrame(np.random.rand(10) for x in range(0, 10)).astype(np.float16).round(2)

Or

np.round(pd.DataFrame(np.random.rand(10) for x in range(0, 10)).astype(np.float16), 2)

Or

pd.DataFrame(np.random.rand(10) for x in range(0, 10)).astype(np.float16).round({0:2, 1:2})

This must have come up before but I can't find it anywhere?

Answer 1

It is rounding. Up to the limits of float16 precision, the results are exactly what you asked for. However, the limits of float16 precision are significantly lower than the 6 significant figures Pandas attempts to print by default, so you see some of the representation imprecision that is usually hidden when printing floating-point numbers.