推理输入契约

导入当前步骤需要的数值计算、预处理、模型或评价工具。依赖集中写在代码开头，便于复现。

import numpy as np
import pandas as pd

单位与缺失策略是业务契约，不能从 dtype 自动推断。

EXPECTED = ["age_years", "amount_yuan"]
UNITS = {"age_years": "years", "amount_yuan": "yuan"}

执行当前代码段，并把得到的状态传给下一步。

def validate(frame, units):
    if frame.columns.duplicated().any() or set(frame.columns) != set(EXPECTED):
        raise ValueError("column mismatch")
    if units != UNITS:
        raise ValueError("unit mismatch")
    aligned = frame.loc[:, EXPECTED]
    if not all(pd.api.types.is_numeric_dtype(dtype) for dtype in aligned.dtypes):
        raise TypeError("numeric columns required")
    if not np.isfinite(aligned.to_numpy(dtype=float)).all():
        raise ValueError("finite values required")
    if not aligned["age_years"].between(0, 120).all():
        raise ValueError("age outside contract")
    if (aligned["amount_yuan"] < 0).any():
        raise ValueError("negative amount")
    return aligned

输出中间参数、形状或最终指标，用于核对代码是否符合预期。

sample = pd.DataFrame({"amount_yuan": [100.0], "age_years": [24]})
validated = validate(sample, UNITS)
print("columns:", validated.columns.tolist())
print("shape:", validated.shape)
try:
    validate(sample, {"age_years": "years", "amount_yuan": "cents"})
except ValueError as error:
    print("rejected:", error)