diff on UInt64 polars series returns series of null instead of actual diff series

[Enaunimes]

diff on UInt64 polars series returns series of null instead of actual diff series

[d-reynol]

@Enaunimes the absolute value of the diff would fit in Uint64, but diffs can be negative & Uint64 is unsigned

[ritchie46]

Because it doesn't fit into i64. Until we have support for i128, there isn't a good fit.

Note that the pandas strategy of converting to float is also wrong:

>>> import pandas as pd
>>> s = [0, 18446743990181366216, 0]
>>> pd.Series(s).diff()
0             NaN
1    1.844674e+19
2    8.352819e+10
dtype: float64

Index 2 is 9 decimals off.

[Enaunimes]

Because it doesn't fit into i64. Until we have support for i128, there isn't a good fit.

Note that the pandas strategy of converting to float is also wrong:

>>> import pandas as pd
>>> s = [0, 18446743990181366216, 0]
>>> pd.Series(s).diff()
0             NaN
1    1.844674e+19
2    8.352819e+10
dtype: float64

Index 2 is 9 decimals off.

I suppose specifying dtype=polars.UInt64 when creating the series should have made the data fit into u64?

[ritchie46]

Sorry, what do you mean?

[Enaunimes]

Sorry, what do you mean?

I'm not from a CS background, so please correct me if I'm wrong.

As in the code sample of the issue, I specified the dtype=polars.UInt64 when creating the series. The number does not fit in the polars.Int64, but should have fit into the polars.Uint64, isn't it?

[Enaunimes]

@Enaunimes the absolute value of the diff would fit in Uint64, but diffs can be negative & Uint64 is unsigned

Got it. Thanks.