fume-manage-python.git

import numpy as np
import pytest
 
import pandas as pd
from pandas import (
    DataFrame,
    Index,
)
import pandas._testing as tm
 
 
@pytest.mark.parametrize(
    "interpolation", ["linear", "lower", "higher", "nearest", "midpoint"]
)
@pytest.mark.parametrize(
    "a_vals,b_vals",
    [
        # Ints
        ([1, 2, 3, 4, 5], [5, 4, 3, 2, 1]),
        ([1, 2, 3, 4], [4, 3, 2, 1]),
        ([1, 2, 3, 4, 5], [4, 3, 2, 1]),
        # Floats
        ([1.0, 2.0, 3.0, 4.0, 5.0], [5.0, 4.0, 3.0, 2.0, 1.0]),
        # Missing data
        ([1.0, np.nan, 3.0, np.nan, 5.0], [5.0, np.nan, 3.0, np.nan, 1.0]),
        ([np.nan, 4.0, np.nan, 2.0, np.nan], [np.nan, 4.0, np.nan, 2.0, np.nan]),
        # Timestamps
        (
            pd.date_range("1/1/18", freq="D", periods=5),
            pd.date_range("1/1/18", freq="D", periods=5)[::-1],
        ),
        (
            pd.date_range("1/1/18", freq="D", periods=5).as_unit("s"),
            pd.date_range("1/1/18", freq="D", periods=5)[::-1].as_unit("s"),
        ),
        # All NA
        ([np.nan] * 5, [np.nan] * 5),
    ],
)
@pytest.mark.parametrize("q", [0, 0.25, 0.5, 0.75, 1])
def test_quantile(interpolation, a_vals, b_vals, q, request):
    if (
        interpolation == "nearest"
        and q == 0.5
        and isinstance(b_vals, list)
        and b_vals == [4, 3, 2, 1]
    ):
        request.node.add_marker(
            pytest.mark.xfail(
                reason="Unclear numpy expectation for nearest "
                "result with equidistant data"
            )
        )
    all_vals = pd.concat([pd.Series(a_vals), pd.Series(b_vals)])
 
    a_expected = pd.Series(a_vals).quantile(q, interpolation=interpolation)
    b_expected = pd.Series(b_vals).quantile(q, interpolation=interpolation)
 
    df = DataFrame({"key": ["a"] * len(a_vals) + ["b"] * len(b_vals), "val": all_vals})
 
    expected = DataFrame(
        [a_expected, b_expected], columns=["val"], index=Index(["a", "b"], name="key")
    )
    if all_vals.dtype.kind == "M" and expected.dtypes.values[0].kind == "M":
        # TODO(non-nano): this should be unnecessary once array_to_datetime
        #  correctly infers non-nano from Timestamp.unit
        expected = expected.astype(all_vals.dtype)
    result = df.groupby("key").quantile(q, interpolation=interpolation)
 
    tm.assert_frame_equal(result, expected)
 
 
def test_quantile_array():
    # https://github.com/pandas-dev/pandas/issues/27526
    df = DataFrame({"A": [0, 1, 2, 3, 4]})
    key = np.array([0, 0, 1, 1, 1], dtype=np.int64)
    result = df.groupby(key).quantile([0.25])
 
    index = pd.MultiIndex.from_product([[0, 1], [0.25]])
    expected = DataFrame({"A": [0.25, 2.50]}, index=index)
    tm.assert_frame_equal(result, expected)
 
    df = DataFrame({"A": [0, 1, 2, 3], "B": [4, 5, 6, 7]})
    index = pd.MultiIndex.from_product([[0, 1], [0.25, 0.75]])
 
    key = np.array([0, 0, 1, 1], dtype=np.int64)
    result = df.groupby(key).quantile([0.25, 0.75])
    expected = DataFrame(
        {"A": [0.25, 0.75, 2.25, 2.75], "B": [4.25, 4.75, 6.25, 6.75]}, index=index
    )
    tm.assert_frame_equal(result, expected)
 
 
def test_quantile_array2():
    # https://github.com/pandas-dev/pandas/pull/28085#issuecomment-524066959
    arr = np.random.RandomState(0).randint(0, 5, size=(10, 3), dtype=np.int64)
    df = DataFrame(arr, columns=list("ABC"))
    result = df.groupby("A").quantile([0.3, 0.7])
    expected = DataFrame(
        {
            "B": [0.9, 2.1, 2.2, 3.4, 1.6, 2.4, 2.3, 2.7, 0.0, 0.0],
            "C": [1.2, 2.8, 1.8, 3.0, 0.0, 0.0, 1.9, 3.1, 3.0, 3.0],
        },
        index=pd.MultiIndex.from_product(
            [[0, 1, 2, 3, 4], [0.3, 0.7]], names=["A", None]
        ),
    )
    tm.assert_frame_equal(result, expected)
 
 
def test_quantile_array_no_sort():
    df = DataFrame({"A": [0, 1, 2], "B": [3, 4, 5]})
    key = np.array([1, 0, 1], dtype=np.int64)
    result = df.groupby(key, sort=False).quantile([0.25, 0.5, 0.75])
    expected = DataFrame(
        {"A": [0.5, 1.0, 1.5, 1.0, 1.0, 1.0], "B": [3.5, 4.0, 4.5, 4.0, 4.0, 4.0]},
        index=pd.MultiIndex.from_product([[1, 0], [0.25, 0.5, 0.75]]),
    )
    tm.assert_frame_equal(result, expected)
 
    result = df.groupby(key, sort=False).quantile([0.75, 0.25])
    expected = DataFrame(
        {"A": [1.5, 0.5, 1.0, 1.0], "B": [4.5, 3.5, 4.0, 4.0]},
        index=pd.MultiIndex.from_product([[1, 0], [0.75, 0.25]]),
    )
    tm.assert_frame_equal(result, expected)
 
 
def test_quantile_array_multiple_levels():
    df = DataFrame(
        {"A": [0, 1, 2], "B": [3, 4, 5], "c": ["a", "a", "a"], "d": ["a", "a", "b"]}
    )
    result = df.groupby(["c", "d"]).quantile([0.25, 0.75])
    index = pd.MultiIndex.from_tuples(
        [("a", "a", 0.25), ("a", "a", 0.75), ("a", "b", 0.25), ("a", "b", 0.75)],
        names=["c", "d", None],
    )
    expected = DataFrame(
        {"A": [0.25, 0.75, 2.0, 2.0], "B": [3.25, 3.75, 5.0, 5.0]}, index=index
    )
    tm.assert_frame_equal(result, expected)
 
 
@pytest.mark.parametrize("frame_size", [(2, 3), (100, 10)])
@pytest.mark.parametrize("groupby", [[0], [0, 1]])
@pytest.mark.parametrize("q", [[0.5, 0.6]])
def test_groupby_quantile_with_arraylike_q_and_int_columns(frame_size, groupby, q):
    # GH30289
    nrow, ncol = frame_size
    df = DataFrame(np.array([ncol * [_ % 4] for _ in range(nrow)]), columns=range(ncol))
 
    idx_levels = [np.arange(min(nrow, 4))] * len(groupby) + [q]
    idx_codes = [[x for x in range(min(nrow, 4)) for _ in q]] * len(groupby) + [
        list(range(len(q))) * min(nrow, 4)
    ]
    expected_index = pd.MultiIndex(
        levels=idx_levels, codes=idx_codes, names=groupby + [None]
    )
    expected_values = [
        [float(x)] * (ncol - len(groupby)) for x in range(min(nrow, 4)) for _ in q
    ]
    expected_columns = [x for x in range(ncol) if x not in groupby]
    expected = DataFrame(
        expected_values, index=expected_index, columns=expected_columns
    )
    result = df.groupby(groupby).quantile(q)
 
    tm.assert_frame_equal(result, expected)
 
 
def test_quantile_raises():
    df = DataFrame([["foo", "a"], ["foo", "b"], ["foo", "c"]], columns=["key", "val"])
 
    with pytest.raises(TypeError, match="cannot be performed against 'object' dtypes"):
        df.groupby("key").quantile()
 
 
def test_quantile_out_of_bounds_q_raises():
    # https://github.com/pandas-dev/pandas/issues/27470
    df = DataFrame({"a": [0, 0, 0, 1, 1, 1], "b": range(6)})
    g = df.groupby([0, 0, 0, 1, 1, 1])
    with pytest.raises(ValueError, match="Got '50.0' instead"):
        g.quantile(50)
 
    with pytest.raises(ValueError, match="Got '-1.0' instead"):
        g.quantile(-1)
 
 
def test_quantile_missing_group_values_no_segfaults():
    # GH 28662
    data = np.array([1.0, np.nan, 1.0])
    df = DataFrame({"key": data, "val": range(3)})
 
    # Random segfaults; would have been guaranteed in loop
    grp = df.groupby("key")
    for _ in range(100):
        grp.quantile()
 
 
@pytest.mark.parametrize(
    "key, val, expected_key, expected_val",
    [
        ([1.0, np.nan, 3.0, np.nan], range(4), [1.0, 3.0], [0.0, 2.0]),
        ([1.0, np.nan, 2.0, 2.0], range(4), [1.0, 2.0], [0.0, 2.5]),
        (["a", "b", "b", np.nan], range(4), ["a", "b"], [0, 1.5]),
        ([0], [42], [0], [42.0]),
        ([], [], np.array([], dtype="float64"), np.array([], dtype="float64")),
    ],
)
def test_quantile_missing_group_values_correct_results(
    key, val, expected_key, expected_val
):
    # GH 28662, GH 33200, GH 33569
    df = DataFrame({"key": key, "val": val})
 
    expected = DataFrame(
        expected_val, index=Index(expected_key, name="key"), columns=["val"]
    )
 
    grp = df.groupby("key")
 
    result = grp.quantile(0.5)
    tm.assert_frame_equal(result, expected)
 
    result = grp.quantile()
    tm.assert_frame_equal(result, expected)
 
 
@pytest.mark.parametrize(
    "values",
    [
        pd.array([1, 0, None] * 2, dtype="Int64"),
        pd.array([True, False, None] * 2, dtype="boolean"),
    ],
)
@pytest.mark.parametrize("q", [0.5, [0.0, 0.5, 1.0]])
def test_groupby_quantile_nullable_array(values, q):
    # https://github.com/pandas-dev/pandas/issues/33136
    df = DataFrame({"a": ["x"] * 3 + ["y"] * 3, "b": values})
    result = df.groupby("a")["b"].quantile(q)
 
    if isinstance(q, list):
        idx = pd.MultiIndex.from_product((["x", "y"], q), names=["a", None])
        true_quantiles = [0.0, 0.5, 1.0]
    else:
        idx = Index(["x", "y"], name="a")
        true_quantiles = [0.5]
 
    expected = pd.Series(true_quantiles * 2, index=idx, name="b", dtype="Float64")
    tm.assert_series_equal(result, expected)
 
 
@pytest.mark.parametrize("q", [0.5, [0.0, 0.5, 1.0]])
@pytest.mark.parametrize("numeric_only", [True, False])
def test_groupby_quantile_raises_on_invalid_dtype(q, numeric_only):
    df = DataFrame({"a": [1], "b": [2.0], "c": ["x"]})
    if numeric_only:
        result = df.groupby("a").quantile(q, numeric_only=numeric_only)
        expected = df.groupby("a")[["b"]].quantile(q)
        tm.assert_frame_equal(result, expected)
    else:
        with pytest.raises(
            TypeError, match="'quantile' cannot be performed against 'object' dtypes!"
        ):
            df.groupby("a").quantile(q, numeric_only=numeric_only)
 
 
def test_groupby_quantile_NA_float(any_float_dtype):
    # GH#42849
    df = DataFrame({"x": [1, 1], "y": [0.2, np.nan]}, dtype=any_float_dtype)
    result = df.groupby("x")["y"].quantile(0.5)
    exp_index = Index([1.0], dtype=any_float_dtype, name="x")
 
    if any_float_dtype in ["Float32", "Float64"]:
        expected_dtype = any_float_dtype
    else:
        expected_dtype = None
 
    expected = pd.Series([0.2], dtype=expected_dtype, index=exp_index, name="y")
    tm.assert_series_equal(result, expected)
 
    result = df.groupby("x")["y"].quantile([0.5, 0.75])
    expected = pd.Series(
        [0.2] * 2,
        index=pd.MultiIndex.from_product((exp_index, [0.5, 0.75]), names=["x", None]),
        name="y",
        dtype=expected_dtype,
    )
    tm.assert_series_equal(result, expected)
 
 
def test_groupby_quantile_NA_int(any_int_ea_dtype):
    # GH#42849
    df = DataFrame({"x": [1, 1], "y": [2, 5]}, dtype=any_int_ea_dtype)
    result = df.groupby("x")["y"].quantile(0.5)
    expected = pd.Series(
        [3.5],
        dtype="Float64",
        index=Index([1], name="x", dtype=any_int_ea_dtype),
        name="y",
    )
    tm.assert_series_equal(expected, result)
 
    result = df.groupby("x").quantile(0.5)
    expected = DataFrame(
        {"y": 3.5}, dtype="Float64", index=Index([1], name="x", dtype=any_int_ea_dtype)
    )
    tm.assert_frame_equal(result, expected)
 
 
@pytest.mark.parametrize(
    "interpolation, val1, val2", [("lower", 2, 2), ("higher", 2, 3), ("nearest", 2, 2)]
)
def test_groupby_quantile_all_na_group_masked(
    interpolation, val1, val2, any_numeric_ea_dtype
):
    # GH#37493
    df = DataFrame(
        {"a": [1, 1, 1, 2], "b": [1, 2, 3, pd.NA]}, dtype=any_numeric_ea_dtype
    )
    result = df.groupby("a").quantile(q=[0.5, 0.7], interpolation=interpolation)
    expected = DataFrame(
        {"b": [val1, val2, pd.NA, pd.NA]},
        dtype=any_numeric_ea_dtype,
        index=pd.MultiIndex.from_arrays(
            [pd.Series([1, 1, 2, 2], dtype=any_numeric_ea_dtype), [0.5, 0.7, 0.5, 0.7]],
            names=["a", None],
        ),
    )
    tm.assert_frame_equal(result, expected)
 
 
@pytest.mark.parametrize("interpolation", ["midpoint", "linear"])
def test_groupby_quantile_all_na_group_masked_interp(
    interpolation, any_numeric_ea_dtype
):
    # GH#37493
    df = DataFrame(
        {"a": [1, 1, 1, 2], "b": [1, 2, 3, pd.NA]}, dtype=any_numeric_ea_dtype
    )
    result = df.groupby("a").quantile(q=[0.5, 0.75], interpolation=interpolation)
 
    if any_numeric_ea_dtype == "Float32":
        expected_dtype = any_numeric_ea_dtype
    else:
        expected_dtype = "Float64"
 
    expected = DataFrame(
        {"b": [2.0, 2.5, pd.NA, pd.NA]},
        dtype=expected_dtype,
        index=pd.MultiIndex.from_arrays(
            [
                pd.Series([1, 1, 2, 2], dtype=any_numeric_ea_dtype),
                [0.5, 0.75, 0.5, 0.75],
            ],
            names=["a", None],
        ),
    )
    tm.assert_frame_equal(result, expected)
 
 
@pytest.mark.parametrize("dtype", ["Float64", "Float32"])
def test_groupby_quantile_allNA_column(dtype):
    # GH#42849
    df = DataFrame({"x": [1, 1], "y": [pd.NA] * 2}, dtype=dtype)
    result = df.groupby("x")["y"].quantile(0.5)
    expected = pd.Series(
        [np.nan], dtype=dtype, index=Index([1.0], dtype=dtype), name="y"
    )
    expected.index.name = "x"
    tm.assert_series_equal(expected, result)
 
 
def test_groupby_timedelta_quantile():
    # GH: 29485
    df = DataFrame(
        {"value": pd.to_timedelta(np.arange(4), unit="s"), "group": [1, 1, 2, 2]}
    )
    result = df.groupby("group").quantile(0.99)
    expected = DataFrame(
        {
            "value": [
                pd.Timedelta("0 days 00:00:00.990000"),
                pd.Timedelta("0 days 00:00:02.990000"),
            ]
        },
        index=Index([1, 2], name="group"),
    )
    tm.assert_frame_equal(result, expected)
 
 
def test_columns_groupby_quantile():
    # GH 33795
    df = DataFrame(
        np.arange(12).reshape(3, -1),
        index=list("XYZ"),
        columns=pd.Series(list("ABAB"), name="col"),
    )
    result = df.groupby("col", axis=1).quantile(q=[0.8, 0.2])
    expected = DataFrame(
        [
            [1.6, 0.4, 2.6, 1.4],
            [5.6, 4.4, 6.6, 5.4],
            [9.6, 8.4, 10.6, 9.4],
        ],
        index=list("XYZ"),
        columns=pd.MultiIndex.from_tuples(
            [("A", 0.8), ("A", 0.2), ("B", 0.8), ("B", 0.2)], names=["col", None]
        ),
    )
 
    tm.assert_frame_equal(result, expected)
 
 
def test_timestamp_groupby_quantile():
    # GH 33168
    df = DataFrame(
        {
            "timestamp": pd.date_range(
                start="2020-04-19 00:00:00", freq="1T", periods=100, tz="UTC"
            ).floor("1H"),
            "category": list(range(1, 101)),
            "value": list(range(101, 201)),
        }
    )
 
    result = df.groupby("timestamp").quantile([0.2, 0.8])
 
    expected = DataFrame(
        [
            {"category": 12.8, "value": 112.8},
            {"category": 48.2, "value": 148.2},
            {"category": 68.8, "value": 168.8},
            {"category": 92.2, "value": 192.2},
        ],
        index=pd.MultiIndex.from_tuples(
            [
                (pd.Timestamp("2020-04-19 00:00:00+00:00"), 0.2),
                (pd.Timestamp("2020-04-19 00:00:00+00:00"), 0.8),
                (pd.Timestamp("2020-04-19 01:00:00+00:00"), 0.2),
                (pd.Timestamp("2020-04-19 01:00:00+00:00"), 0.8),
            ],
            names=("timestamp", None),
        ),
    )
 
    tm.assert_frame_equal(result, expected)
 
 
def test_groupby_quantile_dt64tz_period():
    # GH#51373
    dti = pd.date_range("2016-01-01", periods=1000)
    ser = pd.Series(dti)
    df = ser.to_frame()
    df[1] = dti.tz_localize("US/Pacific")
    df[2] = dti.to_period("D")
    df[3] = dti - dti[0]
    df.iloc[-1] = pd.NaT
 
    by = np.tile(np.arange(5), 200)
    gb = df.groupby(by)
 
    result = gb.quantile(0.5)
 
    # Check that we match the group-by-group result
    exp = {i: df.iloc[i::5].quantile(0.5) for i in range(5)}
    expected = DataFrame(exp).T
    expected.index = expected.index.astype(np.int_)
 
    tm.assert_frame_equal(result, expected)