Property Setter for Subclass of Pandas DataFrame

Question

I am trying to set up a subclass of pd.DataFrame that has two required arguments when initializing (group and timestamp_col). I want to run validation on those arguments group and timestamp_col, so I have a setter method for each of the properties. This all works until I try to set_index() and get TypeError: 'NoneType' object is not iterable. It appears no argument is being passed to my setter function in test_set_index and test_assignment_with_indexed_obj. If I add if g == None: return to my setter function, I can pass the test cases but don't think that is the proper solution.

How should I implement property validation for these required arguments?

Below is my class:

import pandas as pd
import numpy as np


class HistDollarGains(pd.DataFrame):
    @property
    def _constructor(self):
        return HistDollarGains._internal_ctor

    _metadata = ["group", "timestamp_col", "_group", "_timestamp_col"]

    @classmethod
    def _internal_ctor(cls, *args, **kwargs):
        kwargs["group"] = None
        kwargs["timestamp_col"] = None
        return cls(*args, **kwargs)

    def __init__(
        self,
        data,
        group,
        timestamp_col,
        index=None,
        columns=None,
        dtype=None,
        copy=True,
    ):
        super(HistDollarGains, self).__init__(
            data=data, index=index, columns=columns, dtype=dtype, copy=copy
        )

        self.group = group
        self.timestamp_col = timestamp_col

    @property
    def group(self):
        return self._group

    @group.setter
    def group(self, g):
        if g == None:
            return

        if isinstance(g, str):
            group_list = [g]
        else:
            group_list = g

        if not set(group_list).issubset(self.columns):
            raise ValueError("Data does not contain " + '[' + ', '.join(group_list) + ']')
        self._group = group_list

    @property
    def timestamp_col(self):
        return self._timestamp_col

    @timestamp_col.setter
    def timestamp_col(self, t):
        if t == None:
            return
        if not t in self.columns:
            raise ValueError("Data does not contain " + '[' + t + ']')
        self._timestamp_col = t

Here are my test cases:

import pytest

import pandas as pd
import numpy as np

from myclass import *


@pytest.fixture(scope="module")
def sample():
    samp = pd.DataFrame(
        [
            {"timestamp": "2020-01-01", "group": "a", "dollar_gains": 100},
            {"timestamp": "2020-01-01", "group": "b", "dollar_gains": 100},
            {"timestamp": "2020-01-01", "group": "c", "dollar_gains": 110},
            {"timestamp": "2020-01-01", "group": "a", "dollar_gains": 110},
            {"timestamp": "2020-01-01", "group": "b", "dollar_gains": 90},
            {"timestamp": "2020-01-01", "group": "d", "dollar_gains": 100},
        ]
    )

    return samp

@pytest.fixture(scope="module")
def sample_obj(sample):
    return HistDollarGains(sample, "group", "timestamp")

def test_constructor_without_args(sample):
    with pytest.raises(TypeError):
        HistDollarGains(sample)


def test_constructor_with_string_group(sample):
    hist_dg = HistDollarGains(sample, "group", "timestamp")
    assert hist_dg.group == ["group"]
    assert hist_dg.timestamp_col == "timestamp"


def test_constructor_with_list_group(sample):
    hist_dg = HistDollarGains(sample, ["group", "timestamp"], "timestamp")

def test_constructor_with_invalid_group(sample):
    with pytest.raises(ValueError):
        HistDollarGains(sample, "invalid_group", np.random.choice(sample.columns))

def test_constructor_with_invalid_timestamp(sample):
    with pytest.raises(ValueError):
        HistDollarGains(sample, np.random.choice(sample.columns), "invalid_timestamp")

def test_assignment_with_indexed_obj(sample_obj):
    b = sample_obj.set_index(sample_obj.group + [sample_obj.timestamp_col])

def test_set_index(sample_obj):
    # print(isinstance(a, pd.DataFrame))
    assert sample_obj.set_index(sample_obj.group + [sample_obj.timestamp_col]).index.names == ['group', 'timestamp']

Answer 1

The set_index() method will call self.copy() internally to create a copy of your DataFrame object (see the source code here), inside which it uses your customized constructor method, _internal_ctor(), to create the new object (source). Note that self._constructor() is identical to self._internal_ctor(), which is a common internal method for nearly all pandas classes for creating new instances during operations like deep-copy or slicing. Your problem actually originates from this function:

class HistDollarGains(pd.DataFrame):
    ...
    @classmethod
    def _internal_ctor(cls, *args, **kwargs):
        kwargs["group"]         = None
        kwargs["timestamp_col"] = None
        return cls(*args, **kwargs) # this is equivalent to calling
                                    # HistDollarGains(data, group=None, timestamp_col=None)

I guess you copied this code from the github issue. The lines kwargs["**"] = None explicitly tells the constructor to set None to both group and timestamp_col. Finally the setter/validator gets None as the new value and raise an error.

Therefore, you should set an acceptable value to group and timestamp_col.

    @classmethod
    def _internal_ctor(cls, *args, **kwargs):
        kwargs["group"]         = []
        kwargs["timestamp_col"] = 'timestamp' # or whatever name that makes your validator happy
        return cls(*args, **kwargs)

Then you can delete the if g == None: return lines in the validator.