guzhujushi
5749e50e83
实现收尾（Codex 代写，用户授权）： - impute_missing 的 mode 分支改为返回新表（原先就地改调用方的 DataFrame，另外四支都返回新对象） - detect_outliers 补入参类型校验（§4.7：非法类型抛 TypeError，不静默兜底）： 接受 DataFrame / Series / list / tuple / ndarray；tuple 同时补进 is_ld_array； dict 不再被接受（以前靠 pd.DataFrame() 碰巧能过，见决策 33） 测试： - 摘掉最后一条 xfail（mode 就地改）并加强断言 - 新增 3 条：非法类型抛 TypeError、元组按一维数组型处理、全 NaN 列与单元素列不抛异常 - 64 passed / 42 skipped / 0 xfailed（xfail 清零，7 条契约全部转正）；ruff 0；pyright 0 文档： - 新增 docs/algorithms/detect_outliers.md —— 第一张 S 档函数卡片（roadmap 要求约 15 张） 除模板九节外多记三条判断：为什么不返回 Result、非数值列恒 False 的两条备选路各错在哪、 ddof 口径与 stats.zscore 不一致（前者 1、后者 0，小样本差一成） - 新增 examples/01_clean_and_describe.py —— 六个 data 函数的最小用法，实测跑通（exit 0） - STATUS.md：E 节收尾、验证结果表、下一步重写（transform/standardize 的待定项）、决策 33 - CHANGELOG.md：把 [Unreleased] 里重复 7 次的 Added/Changed 小节合并成 Added/Changed/Notes； 并修掉上轮遗留的缺陷 —— pyright 那条 bullet 的开头被吞掉，半句挂在 ruff 那条后面
388 lines
13 KiB
Python
388 lines
13 KiB
Python
"""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()) == []
|
||
|
||
|
||
# --------------------------------------------------------------- data: clean
|
||
|
||
|
||
def test_clean_missing_default_drops_any_missing():
|
||
"""不传 threshold 时是"只要有缺失就删"。"""
|
||
from mcm_toolkit.data import clean_missing
|
||
|
||
df = pd.DataFrame({"a": [1.0, np.nan, 3.0], "b": [1.0, 2.0, np.nan]})
|
||
out = clean_missing(df)
|
||
assert len(out) == 1
|
||
assert list(out.columns) == ["a", "b"]
|
||
|
||
|
||
def test_clean_missing_threshold_keeps_rows_within_limit():
|
||
"""四行三列、阈值 0.5：缺 2/3 的两行要删，剩 2 行。
|
||
|
||
3 列时 min_non_na 必须是 ceil(3*0.5)=2；写成 int() 截断就会漏删。
|
||
"""
|
||
from mcm_toolkit.data import clean_missing
|
||
|
||
df = pd.DataFrame(
|
||
{
|
||
"a": [1.0, np.nan, np.nan, 4.0],
|
||
"b": [1.0, np.nan, np.nan, 4.0],
|
||
"c": [1.0, 2.0, np.nan, 4.0],
|
||
}
|
||
)
|
||
out = clean_missing(df, axis=0, threshold=0.5)
|
||
assert len(out) == 2
|
||
assert out["a"].tolist() == [1.0, 4.0]
|
||
|
||
|
||
def test_clean_missing_axis1_drops_empty_column():
|
||
"""axis=1 按列删：整列全空（缺失率 1.0 > 0.5）要掉。"""
|
||
from mcm_toolkit.data import clean_missing
|
||
|
||
df = pd.DataFrame({"x": [1.0, 2.0, 3.0], "y": [np.nan, np.nan, np.nan]})
|
||
assert list(clean_missing(df, axis=1, threshold=0.5).columns) == ["x"]
|
||
|
||
|
||
def test_clean_missing_rejects_bad_axis():
|
||
from mcm_toolkit.data import clean_missing
|
||
|
||
with pytest.raises(ValueError, match="axis"):
|
||
clean_missing(pd.DataFrame({"a": [1.0]}), axis=2)
|
||
|
||
|
||
def test_clean_missing_rejects_non_dataframe():
|
||
from mcm_toolkit.data import clean_missing
|
||
|
||
with pytest.raises(TypeError):
|
||
clean_missing([[1.0, 2.0]])
|
||
|
||
|
||
def test_impute_missing_mean_and_ffill():
|
||
from mcm_toolkit.data import impute_missing
|
||
|
||
df = pd.DataFrame({"a": [1.0, np.nan, 3.0]})
|
||
assert impute_missing(df, method="mean")["a"].tolist() == [1.0, 2.0, 3.0]
|
||
assert impute_missing(df, method="ffill")["a"].tolist() == [1.0, 1.0, 3.0]
|
||
|
||
|
||
def test_impute_missing_mode_survives_all_nan_column():
|
||
"""整列全 NaN 时 mode() 给不出众数：不能崩，也不能硬填一个值进去。"""
|
||
from mcm_toolkit.data import impute_missing
|
||
|
||
df = pd.DataFrame({"a": [np.nan, np.nan], "b": [1.0, np.nan]})
|
||
out = impute_missing(df, method="mode")
|
||
assert bool(out["a"].isna().all())
|
||
assert out["b"].tolist() == [1.0, 1.0]
|
||
|
||
|
||
def test_impute_missing_rejects_unknown_method():
|
||
from mcm_toolkit.data import impute_missing
|
||
|
||
with pytest.raises(ValueError, match="bogus"):
|
||
impute_missing(pd.DataFrame({"a": [1.0]}), method="bogus")
|
||
|
||
|
||
def test_impute_missing_rejects_non_dataframe():
|
||
from mcm_toolkit.data import impute_missing
|
||
|
||
with pytest.raises(TypeError):
|
||
impute_missing([[1.0, 2.0]])
|
||
|
||
|
||
def test_winsorize_clips_series_to_quantiles():
|
||
from mcm_toolkit.data import winsorize
|
||
|
||
s = pd.Series([1.0, 2.0, 3.0, 4.0, 1000.0])
|
||
out = winsorize(s, lower=0.1, upper=0.9)
|
||
assert float(out.iloc[0]) == pytest.approx(1.4)
|
||
assert float(out.iloc[-1]) == pytest.approx(601.6)
|
||
assert float(out.max()) < 1000.0
|
||
|
||
|
||
def test_winsorize_rejects_bad_quantile_order():
|
||
from mcm_toolkit.data import winsorize
|
||
|
||
with pytest.raises(ValueError, match="lower"):
|
||
winsorize(pd.Series([1.0, 2.0]), lower=0.9, upper=0.1)
|
||
|
||
|
||
def test_winsorize_rejects_non_frame_with_typeerror():
|
||
"""§4.7：类型不对必须抛 TypeError，不能抛 ValueError。"""
|
||
from mcm_toolkit.data import winsorize
|
||
|
||
with pytest.raises(TypeError):
|
||
winsorize([1.0, 2.0, 3.0])
|
||
|
||
|
||
def test_detect_outliers_rejects_unknown_method():
|
||
"""方法名校验在取默认取值之前，所以这条现在就是绿的。"""
|
||
from mcm_toolkit.data import detect_outliers
|
||
|
||
with pytest.raises(ValueError, match="bogus"):
|
||
detect_outliers([1.0, 2.0], method="bogus") # type: ignore[arg-type]
|
||
|
||
|
||
def test_detect_outliers_rejects_bad_type():
|
||
"""§4.7：类型不对必须抛 TypeError，不能让它一路飘进 pandas 里去。"""
|
||
from mcm_toolkit.data import detect_outliers
|
||
|
||
for bad in (42, 3.14, "123", None, {"a": [1, 2]}):
|
||
with pytest.raises(TypeError):
|
||
detect_outliers(bad)
|
||
|
||
|
||
def test_detect_outliers_survives_degenerate_columns():
|
||
"""全 NaN 列与单元素列都不能抛异常，判定一律为 False。"""
|
||
from mcm_toolkit.data import detect_outliers
|
||
|
||
df = pd.DataFrame({"all_nan": [np.nan, np.nan, np.nan], "one": [1.0, 2.0, 3.0]})
|
||
assert not bool(np.asarray(detect_outliers(df)).any())
|
||
assert not bool(np.asarray(detect_outliers([5.0])).any())
|
||
assert not bool(np.asarray(detect_outliers([5.0], method="zscore")).any())
|
||
|
||
|
||
def test_detect_outliers_accepts_tuple_as_1d():
|
||
"""元组按一维数组型处理，返回值与传 list 一致。"""
|
||
from mcm_toolkit.data import detect_outliers
|
||
|
||
mask = np.asarray(detect_outliers((1.0, 2.0, 3.0, 4.0, 1000.0)))
|
||
assert mask.shape == (5,)
|
||
assert mask.dtype == bool
|
||
assert bool(mask[-1])
|
||
|
||
|
||
# ------------------------ 曾经 xfail 的契约，现已全部实现（留作回归）
|
||
#
|
||
# 这些用例写的是**文档里已经承诺的行为**。实现没跟上时用 xfail，让它们照样出现在
|
||
# 报告里（显示成 x）；修好之后会自动变成 XPASS —— 看到 XPASS 就把 mark 摘掉。
|
||
#
|
||
# 已摘掉：threshold 范围校验、IQR 的 UnboundLocalError、(n,1) 形状、zscore 默认阈值、
|
||
# winsorize 的文本列、detect_outliers 的文本列、impute_missing 的 mode 就地修改。
|
||
# **当前 0 条 xfail** —— 以后新增契约照旧：先写测试 + xfail，实现跟上就摘 mark。
|
||
|
||
|
||
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 int(mask.sum()) == 3
|
||
assert bool(mask[-3:].all())
|
||
|
||
|
||
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
|
||
|
||
|
||
def test_detect_outliers_zscore_flags_planted_point():
|
||
from mcm_toolkit.data import detect_outliers
|
||
|
||
x = np.concatenate([np.zeros(99), [1e6]])
|
||
mask = np.asarray(detect_outliers(x, method="zscore"))
|
||
assert int(mask.sum()) == 1
|
||
assert bool(mask[-1].all())
|
||
|
||
|
||
def test_detect_outliers_survives_text_columns():
|
||
"""赛题表几乎都带文本列：分位数只在数值列上算，文本列恒 False。"""
|
||
from mcm_toolkit.data import detect_outliers
|
||
|
||
df = pd.DataFrame({"city": ["a", "b", "c", "d"], "v": [1.0, 2.0, 3.0, 1000.0]})
|
||
mask = detect_outliers(df)
|
||
assert int(np.asarray(mask).sum()) == 1
|
||
assert not bool(np.asarray(mask["city"]).any())
|
||
assert bool(np.asarray(mask["v"]).any())
|
||
|
||
|
||
def test_detect_outliers_all_text_frame_returns_all_false():
|
||
"""整表没有数值列时不报错，返回全 False 掩码（语义：这一列没有离群点）。"""
|
||
from mcm_toolkit.data import detect_outliers
|
||
|
||
df = pd.DataFrame({"a": ["x", "y"], "b": ["z", "w"]})
|
||
mask = detect_outliers(df)
|
||
assert mask.shape == df.shape
|
||
assert not bool(np.asarray(mask).any())
|
||
|
||
|
||
def test_winsorize_survives_text_columns():
|
||
"""文本列必须原样留着：分位数只在数值列上算。"""
|
||
from mcm_toolkit.data import winsorize
|
||
|
||
df = pd.DataFrame({"city": ["a", "b", "c", "d"], "v": [1.0, 2.0, 3.0, 1000.0]})
|
||
out = winsorize(df)
|
||
assert out["city"].tolist() == ["a", "b", "c", "d"]
|
||
assert float(out["v"].iloc[-1]) < 1000.0
|
||
|
||
|
||
def test_clean_missing_rejects_out_of_range_threshold():
|
||
from mcm_toolkit.data import clean_missing
|
||
|
||
with pytest.raises(ValueError, match="threshold"):
|
||
clean_missing(pd.DataFrame({"a": [1.0, np.nan]}), threshold=1.5)
|
||
|
||
|
||
def test_impute_missing_mode_does_not_mutate_input():
|
||
"""mode 分支也必须返回新表：不能顺手把调用方的表改了。"""
|
||
from mcm_toolkit.data import impute_missing
|
||
|
||
df = pd.DataFrame({"a": [1.0, np.nan, 1.0]})
|
||
out = impute_missing(df, method="mode")
|
||
assert out["a"].tolist() == [1.0, 1.0, 1.0]
|
||
assert bool(df["a"].isna().any())
|
||
|
||
|
||
# ------------------------------------------------- 以下为未实现算法（L0/L1 待写）
|
||
|
||
|
||
@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
|