快速开始

仓库提供了可直接运行的完整示例：

python examples/00_quickstart.py

下面的 Python 代码块按顺序执行即可完成同一条基础流程。示例只使用基础依赖。

1. 准备数据并按时间切分

import numpy as np
import pandas as pd
from sklearn.datasets import make_classification

import hscredit
from hscredit.core.model_selection import time_train_test_split

X_array, y_array = make_classification(
    n_samples=400,
    n_features=4,
    n_informative=3,
    n_redundant=0,
    weights=[0.72, 0.28],
    random_state=42,
)
df = pd.DataFrame(X_array, columns=["age", "income", "debt_ratio", "apply_count"])
df["apply_date"] = pd.date_range("2024-01-01", periods=len(df), freq="D")
df["customer_id"] = [f"C{i // 2:04d}" for i in range(len(df))]
df["target"] = y_array

train_df, test_df = time_train_test_split(df, "apply_date", test_size=0.25)
features = ["age", "income", "debt_ratio", "apply_count"]
X_train, y_train = train_df[features], train_df["target"]
X_test, y_test = test_df[features], test_df["target"]

2. 数据探索

import hscredit.core.eda as eda

summary = eda.data_info(train_df)
iv_result = eda.batch_iv_analysis(train_df, features=features, target="target")
trend = eda.bad_rate_trend(train_df, target_col="target", date_col="apply_date")

3. 分箱与变量筛选

from hscredit.core.binning import OptimalBinning
from hscredit.core.selectors import IVSelector, VIFSelector, CompositeFeatureSelector

binner = OptimalBinning(method="best_iv", max_n_bins=5)
binner.fit(X_train, y_train)
X_train_woe = binner.transform(X_train, metric="woe")

selector = CompositeFeatureSelector([
    ("iv", IVSelector(threshold=0.0)),
    ("vif", VIFSelector(threshold=20.0)),
])
X_selected = selector.fit_transform(X_train_woe, y_train)

4. 评分卡建模

from hscredit.core.models import ScoreCard

scorecard = ScoreCard(
    pdo=60,
    rate=2,
    base_odds=35,
    base_score=750,
    binner=binner,
)
scorecard.fit(X_train_woe, y_train)
scores = scorecard.predict(X_test)

5. 机器学习模型与概率校准

from sklearn.model_selection import train_test_split

from hscredit.core.models import DecisionTreeClassifier, RandomForest, SVM
from hscredit.core.models.calibration import ProbabilityCalibrator

X_model_train, X_calib, y_model_train, y_calib = train_test_split(
    X_train,
    y_train,
    test_size=0.2,
    random_state=42,
    stratify=y_train,
)

model = RandomForest(n_estimators=30, random_state=42)
model.fit(X_model_train, y_model_train)
metrics = model.evaluate(X_test, y_test)

calibrator = ProbabilityCalibrator(
    model=model,
    method="platt",
    calib_ratio=None,
).fit(X_calib, y_calib)
calibrated_proba = calibrator.predict_proba(X_test)
calibration_report = calibrator.report(X_test, y_test)

svm_model = SVM(C=1.0, kernel="rbf", random_state=42).fit(X_train, y_train)
tree_model = DecisionTreeClassifier(max_depth=4, random_state=42).fit(X_train, y_train)
svm_proba = svm_model.predict_proba(X_test)[:, 1]
tree_proba = tree_model.predict_proba(X_test)[:, 1]

6. 策略规则挖掘

from hscredit.report.mining import SingleFeatureRuleMiner, TreeRuleExtractor

single_miner = SingleFeatureRuleMiner(
    target="target",
    method="best_iv",
    max_n_bins=4,
    min_samples=5,
).fit(train_df[features + ["target"]])
single_rules = single_miner.get_top_rules(top_n=5, metric="lift")

extractor = TreeRuleExtractor(max_depth=3, min_samples_leaf=10)
extractor.fit(X_train, y_train)
tree_rules = extractor.extract_rules()

7. 模型报告和制品保存

from hscredit.report import auto_model_report

report = auto_model_report(
    model,
    X_train=X_train,
    y_train=y_train,
    X_test=X_test,
    y_test=y_test,
    method="predict_score",
    excel_path="模型评估报告.xlsx",
    verbose=False,
    with_plots=False,
)

model.save_artifact("risk_model.joblib")
restored_model = RandomForest.load_artifact("risk_model.joblib")
assert restored_model.scorecard_.transformer_ is restored_model.score_transformer_

模型在 fit 后可通过 model.score_transformer_ 直接访问使用完整训练概率拟合的转换器。sklearn.clone 只保留其初始化配置；copy、joblib/pickle 和 native sidecar 保存拟合状态。

当前版本不再提供七个具体模型的旧 RiskModel 后缀类名。引用旧类路径的 Python 序列化制品应先使用旧版 hscredit 加载并重新训练或导出，再升级使用。

8. 规则表达式

from hscredit.core.rules import Rule

rule = Rule("age >= 0") & Rule("income < 1")
rule_report = rule.report(test_df, target="target")