DataFrame.groupby.first to match pandas

[ajpotts]

def build_ak_df_with_nans():
    data = {
        "key1": ["valuew", "valuex", "valuew", "valuex"],
        "key2": ["valueA", "valueB", "valueA", "valueB"],
        "nums1": [1, np.nan, 3, 4],
        "nums2": [1, np.nan, np.nan, 7],
        "nums3": [10, 8, 9, 7],
        "bools": [True, False, True, False],
    }

    ak_df = ak.DataFrame({k: ak.array(v) for k, v in data.items()})
    return ak_df

df.drop('bools', axis=1, inplace=True)
pd_df = df.to_pandas()

Arkouda gives one result:

In [143]: df.groupby(["key1","key2"]).first()
Out[143]: 
     key1    key2  nums1  nums2  nums3
0  valuew  valueA    1.0    1.0     10
1  valuex  valueB    NaN    NaN      8 (2 rows x 5 columns)

Pandas gives a different result:

In [144]: pd_df.groupby(["key1","key2"], as_index=False).first()
Out[144]: 
     key1    key2  nums1  nums2  nums3
0  valuew  valueA    1.0    1.0     10
1  valuex  valueB    4.0    7.0      8