MCM-ICM_Toolkit/tests/test_data.py

"""data 包测试。

L0 阶段的约定：**只测已实现的行为**。未实现的算法用逐条 skip 标出，
不要用模块级 pytestmark —— 那会把同文件里已实现的函数一起跳过，
让 load_table 这种已经写好的函数变成零覆盖。
"""

from __future__ import annotations

from pathlib import Path

import numpy as np
import pandas as pd
import pytest

_NOT_IMPLEMENTED = "未实现：见 docs/03-roadmap.md L0/L1"


# ---------------------------------------------------------------- io: load_table


def test_load_table_detects_gbk(tmp_path):
    """国内赛题数据常见 GBK，编码探测必须能兜住。"""
    from mcm_toolkit.data import load_table

    p = tmp_path / "gbk.csv"
    p.write_bytes("城市,人口\n北京,2189\n".encode("gbk"))
    df = load_table(p)
    assert list(df.columns) == ["城市", "人口"]


def test_load_table_keeps_utf8_readable(tmp_path):
    """UTF-8 文件必须按 UTF-8 读出，不能先试 GBK 变成乱码。"""
    from mcm_toolkit.data import load_table

    p = tmp_path / "utf8.csv"
    p.write_text("省份,GDP增速\n广东,5.8\n", encoding="utf-8")
    df = load_table(p)
    assert list(df.columns) == ["省份", "GDP增速"]
    assert df.loc[0, "省份"] == "广东"


def test_load_table_maps_na_like_tokens_to_nan(tmp_path):
    """赛题里的缺失值长得五花八门，默认 na_values 要全部吃成 NaN。"""
    from mcm_toolkit.data import load_table

    p = tmp_path / "na.csv"
    p.write_text("a,b,c\n1,NA,\n2,-,#N/A\n", encoding="utf-8")
    df = load_table(p)
    # 外面套 bool()：df["b"] 的类型是 Series | DataFrame，直接当条件用 pyright 会报
    # "条件操作数无效"；套上 bool() 既满足类型检查，又能在它真返回 Series 时当场炸掉。
    assert bool(df["b"].isna().all())
    assert bool(df["c"].isna().all())
    assert bool(df["a"].notna().all())


def test_load_table_respects_explicit_encoding(tmp_path):
    """显式给了 encoding 就照办，不再做探测。"""
    from mcm_toolkit.data import load_table

    p = tmp_path / "gbk.csv"
    p.write_bytes("城市,人口\n上海,2487\n".encode("gbk"))
    df = load_table(p, encoding="gbk")
    assert list(df.columns) == ["城市", "人口"]


def test_load_table_rejects_unknown_suffix(tmp_path):
    """不支持的扩展名要抛 ValueError，而且要带上后缀名。"""
    from mcm_toolkit.data import load_table

    p = tmp_path / "data.txt"
    p.write_text("a,b\n1,2\n", encoding="utf-8")
    with pytest.raises(ValueError, match=r"\.txt"):
        load_table(p)


def test_load_table_missing_file_raises(tmp_path):
    """文件不存在时要把路径原样带出来，方便定位。"""
    from mcm_toolkit.data import load_table

    missing = tmp_path / "nope.csv"
    with pytest.raises(FileNotFoundError, match=r"nope\.csv"):
        load_table(missing)


# ---------------------------------------------------------------- io: save_table


def test_save_table_round_trip(tmp_path):
    """存进去再读出来，内容要一致，并且返回实际写入的 Path。"""
    from mcm_toolkit.data import load_table, save_table

    df = pd.DataFrame({"城市": ["北京", "上海"], "人口": [2189.0, 2487.0]})
    out = save_table(df, tmp_path / "out.csv")
    assert isinstance(out, Path)
    assert out.exists()
    back = load_table(out)
    assert list(back.columns) == ["城市", "人口"]
    assert back["人口"].tolist() == [2189.0, 2487.0]


def test_save_table_rejects_non_dataframe(tmp_path):
    """传错类型要当场报错，不要写出一个奇怪的文件。"""
    from mcm_toolkit.data import save_table

    with pytest.raises(TypeError):
        save_table([1, 2, 3], tmp_path / "bad.csv")  # type: ignore[arg-type]  # 就是要传错


def test_save_table_validates_before_touching_disk(tmp_path):
    """校验必须早于副作用：参数不合法时，一个目录、一个文件都不能留下。"""
    from mcm_toolkit.data import save_table

    with pytest.raises(ValueError, match=r"\.txt"):
        save_table(pd.DataFrame({"a": [1]}), tmp_path / "d1" / "x.txt")
    with pytest.raises(TypeError):
        save_table([1, 2, 3], tmp_path / "d2" / "x.csv")  # type: ignore[arg-type]  # 就是要传错

    assert sorted(p.name for p in tmp_path.iterdir()) == []


# ------------------------------------------------- 以下为未实现算法（L0/L1 待写）


@pytest.mark.skip(reason=_NOT_IMPLEMENTED)
def test_detect_outliers_finds_planted_points():
    """人工埋 3 个离群点，IQR 法应正好命中 3 个。"""
    from mcm_toolkit.data import detect_outliers

    x = np.concatenate([np.arange(100.0), [1e6, -1e6, 1e7]])
    mask = np.asarray(detect_outliers(x, method="iqr"))
    assert mask.sum() == 3
    assert mask[-3:].all()


@pytest.mark.skip(reason=_NOT_IMPLEMENTED)
def test_detect_outliers_returns_mask_not_filtered_data():
    """约定：只标不删——返回与原数据同形的布尔掩码。"""
    from mcm_toolkit.data import detect_outliers

    x = np.arange(50.0)
    mask = np.asarray(detect_outliers(x, method="zscore"))
    assert mask.shape == x.shape
    assert mask.dtype == bool


@pytest.mark.skip(reason=_NOT_IMPLEMENTED)
def test_normalize_minmax_range(rng):
    from mcm_toolkit.data import normalize

    x = rng.normal(size=(50, 3))
    out = np.asarray(normalize(x, method="minmax"))
    assert np.allclose(out.min(axis=0), 0.0)
    assert np.allclose(out.max(axis=0), 1.0)


@pytest.mark.skip(reason=_NOT_IMPLEMENTED)
def test_standardize_has_zero_mean_unit_std(rng):
    from mcm_toolkit.data import standardize

    x = rng.normal(5, 3, size=1000)
    out = np.asarray(standardize(x))
    assert abs(out.mean()) < 1e-9
    assert abs(out.std() - 1.0) < 1e-9


@pytest.mark.skip(reason=_NOT_IMPLEMENTED)
def test_train_test_split_sizes_and_no_overlap(rng):
    from mcm_toolkit.data import train_test_split

    X = rng.normal(size=(100, 2))
    Xtr, Xte = train_test_split(X, test_size=0.2, random_state=0)
    assert len(Xtr) == 80 and len(Xte) == 20
    assert len({tuple(r) for r in Xtr} & {tuple(r) for r in Xte}) == 0