属性 Setter 用于 Pandas DataFrame 的子类

Question

我正在尝试设置 pd.DataFrame 的子 class，它在初始化时有两个必需的参数（group 和 timestamp_col）。我想运行验证这些参数 group 和 timestamp_col，所以我为每个属性设置了一个 setter 方法。这一切都有效，直到我尝试 set_index() 并获得 TypeError: 'NoneType' object is not iterable。在 test_set_index 和 test_assignment_with_indexed_obj 中似乎没有参数被传递到我的 setter 函数。如果我将 if g == None: return 添加到我的 setter 函数，我可以通过测试用例，但我认为这不是正确的解决方案。

我应该如何对这些必需的参数实施属性验证？

下面是我的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

这是我的测试用例：

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

set_index() 方法将在内部调用 self.copy() 以创建 DataFrame 对象的副本（请参阅源代码 here), inside which it uses your customized constructor method, _internal_ctor(), to create the new object (source）。请注意，self._constructor() 与 self._internal_ctor() 相同，这是几乎所有 pandas 类的通用内部方法，用于在深度复制或切片等操作期间创建新实例。你的问题其实来源于这个函数：

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)

我猜你从 the github issue 复制了这段代码。 kwargs["**"] = None 行明确告诉构造函数将 None 设置为 group 和 timestamp_col。最后 setter/validator 获得 None 作为新值并引发错误。

因此，您应该将可接受的值设置为 group 和 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)

然后你可以删除验证器中的if g == None: return行。

属性 Setter 用于 Pandas DataFrame 的子类

Property Setter for Subclass of Pandas DataFrame

python

properties

subclass

pandas