feat(#39): 超参包 JSON Schema 校验器（PRD 5.3 模型框架配置化）
新增 core/model-framework/hyperparam.py：超参包（Hyperparam Pack）结构化 校验器，零外部依赖（不依赖 jsonschema），对齐 PRD 5.3「超参包驱动」与 EPIC #5。 校验维度（10 类）： - 顶层结构：必填字段 model_id/template/algorithm/features/target； - algorithm：已知算法集合校验（xgboost/lightgbm/dnn/lstm/gnn/ isolation_forest/zscore/linear/ridge），未知项报错并列出已知项 （对齐 PRD「新增结构走 Recipe 插件注册」）； - features：非空列表、name 唯一且合法、spec 非空（FeatureSpec 文本 存在性校验，语法由 #35 引擎解释）； - objective/hyperparams/train_window/alarm_threshold/drift_check 可选 字段的类型、取值与格式校验（train_window 形如 180d/4w；drift limit 在 (0,1]；alarm type 含对应阈值键）。 设计： - validate_pack 返回 ValidationReport（逐条问题，不抛异常），便于配置台 聚合展示；load_pack 校验失败抛 ValueError 供训练/推理流水线 fail-fast。 - validate_pack_file 复用结构校验，文件/JSON 解析错误也记入报告。 测试：tests/test_hyperparam.py 25 用例全绿（合法包/必填缺失/未知算法/ 特征重复与缺 spec/目标函数/train_window 格式/alarm/drift/文件加载与 JSON 解析错误）；python -m unittest discover -s tests -v → 25 passed。 close #39
This commit is contained in:
@@ -0,0 +1,37 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""iAOP-Core · AI 模型框架（AI Model Framework）—— 配置驱动的可模板化模型内核。
|
||||
|
||||
对应 PRD 5.3「③ AI 模型框架」与 EPIC #5（平台化改造 RISK 项）：
|
||||
固定主干网络 + 可配置超参；Model Recipe 插件注册；FeatureSpec 声明式特征；
|
||||
所有可变量外置为 **JSON 超参包**，同一框架切换模板仅改此包（零改码）。
|
||||
|
||||
模块组成（按 EPIC #5 拆分的 ≤0.5d 子任务逐步落地）：
|
||||
- hyperparam 超参包 JSON Schema 校验器（Issue #39）：加载 + 校验超参包，
|
||||
返回逐条校验报告，供配置台与训练/推理流水线复用。
|
||||
- （后续）recipe Model Recipe 插件接口与样例协议（Issue #34）
|
||||
- （后续）feature FeatureSpec 声明式特征定义引擎（Issue #35）
|
||||
- （后续）registry 模型模板注册 / 加载 / 版本机制（Issue #41）
|
||||
|
||||
超参包 JSON 结构见 PRD 5.3「超参包 JSON 完整示例」与 ``config/`` 下样例。
|
||||
|
||||
测试：`python -m unittest discover -s tests -v`（在 core/model-framework 目录下执行）。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from .hyperparam import (
|
||||
HyperparamPack,
|
||||
ValidationIssue,
|
||||
ValidationReport,
|
||||
load_pack,
|
||||
validate_pack,
|
||||
validate_pack_file,
|
||||
)
|
||||
|
||||
__all__ = [
|
||||
"HyperparamPack",
|
||||
"ValidationIssue",
|
||||
"ValidationReport",
|
||||
"load_pack",
|
||||
"validate_pack",
|
||||
"validate_pack_file",
|
||||
]
|
||||
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"model_id": "quality_predict_ti",
|
||||
"template": "iAOP-Template-Ti",
|
||||
"algorithm": "xgboost",
|
||||
"features": [
|
||||
{"name": "EMA_CLF_TEMP_5m", "spec": "EMA(CLF-01.TEMP, 5m)"},
|
||||
{"name": "RollingStd_CL2_10", "spec": "RollingStd(CLF-01.CL2, 10)"},
|
||||
{"name": "ROC_FURNACE_PRESS", "spec": "RateOfChange(CLF-01.PRES)"}
|
||||
],
|
||||
"target": "Ti_purity",
|
||||
"objective": "reg:squarederror",
|
||||
"hyperparams": {"max_depth": 6, "eta": 0.1, "n_estimators": 300},
|
||||
"train_window": "180d",
|
||||
"alarm_threshold": {"type": "zscore", "k": 3.0},
|
||||
"drift_check": {"method": "psi", "limit": 0.2}
|
||||
}
|
||||
@@ -0,0 +1,339 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""超参包（Hyperparam Pack）JSON Schema 校验器。
|
||||
|
||||
对应 issue #39（父 EPIC #5「③ AI 模型框架 配置化重构」）与 PRD 5.3
|
||||
「超参包驱动」：所有可变量（输入特征清单、算法选型、超参、训练窗口、告警阈值、
|
||||
目标函数）外置为 JSON 超参包，同一框架切换模板仅改此包。
|
||||
|
||||
本模块对超参包做 **结构化校验**（不依赖 jsonschema 第三方库，零外部依赖，
|
||||
便于边缘 / 离线环境运行），返回逐条问题的校验报告，便于配置台一次性展示全部问题、
|
||||
也便于训练/推理流水线在加载包时 fail-fast。
|
||||
|
||||
校验维度（对齐 PRD 5.3 超参包示例与 5.3「配置点」）：
|
||||
1. 顶层结构：必填字段 model_id / template / algorithm / features / target；
|
||||
2. model_id / template：非空字符串；
|
||||
3. algorithm：必须在合法算法集合内（xgboost / lightgbm / dnn / lstm / gnn /
|
||||
isolation_forest / zscore / linear / ridge）；未知算法报错并提示已知项
|
||||
（对齐 PRD「新增结构走插件注册」——校验期即暴露非法选型）；
|
||||
4. features：非空列表；每项含 name（非空、包内唯一）与 spec（非空字符串，
|
||||
FeatureSpec 文本，由 issue #35 引擎解释，此处只做存在性校验）；
|
||||
5. target：非空字符串；
|
||||
6. objective（可选）：若填写必须为已知目标函数（reg:squarederror /
|
||||
binary:logistic / multi:softmax / regression / classification 等）；
|
||||
7. hyperparams（可选）：若提供必须为对象（dict），且不含空键；
|
||||
8. train_window（可选）：若填写必须形如 ``<正整数><单位>``（单位 d/w/h/m），
|
||||
如 ``180d`` / ``4w``；
|
||||
9. alarm_threshold（可选）：若提供必须为对象，且含 ``type``
|
||||
（zscore / quantile / absolute）与对应阈值键；
|
||||
10. drift_check（可选）：若提供必须为对象，含 ``method``（psi / ks / chi2）
|
||||
与 ``limit``（0~1 之间）。
|
||||
|
||||
注：返回报告而非抛异常，便于配置台聚合展示；``load_pack`` 在校验失败时抛
|
||||
``ValueError`` 供流水线 fail-fast。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Any, Dict, List, Optional, Tuple
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# 合法取值集合（对齐 PRD 5.3「四类模型模板」+ 默认主干网络 + Recipe 插件扩展点）
|
||||
# ---------------------------------------------------------------------------
|
||||
#: 已注册算法（默认主干 + Recipe 可选结构）。新增结构应走插件注册（issue #34/#41），
|
||||
#: 此处枚举的是内核自带项；校验期遇到未知 algorithm 会报错并列出已知项，避免
|
||||
#: 静默落到错误分支。
|
||||
KNOWN_ALGORITHMS: Tuple[str, ...] = (
|
||||
# 监督回归 / 分类（质量预测）
|
||||
"xgboost",
|
||||
"lightgbm",
|
||||
"linear",
|
||||
"ridge",
|
||||
# 神经网络主干（默认 DNN；Recipe 可选 LSTM/GNN）
|
||||
"dnn",
|
||||
"lstm",
|
||||
"gnn",
|
||||
# 无监督异常 / 杂质预警
|
||||
"isolation_forest",
|
||||
"zscore",
|
||||
)
|
||||
|
||||
#: 已知目标函数（xgboost 风格 + 通用风格）。
|
||||
KNOWN_OBJECTIVES: Tuple[str, ...] = (
|
||||
"reg:squarederror",
|
||||
"reg:squaredlogerror",
|
||||
"binary:logistic",
|
||||
"multi:softmax",
|
||||
"multi:softprob",
|
||||
"regression",
|
||||
"classification",
|
||||
)
|
||||
|
||||
#: 训练窗口合法时间单位。
|
||||
_TRAIN_WINDOW_RE = re.compile(r"^\d+(\.\d+)?[dwhm]$")
|
||||
#: FeatureSpec 仅校验非空文本（具体语法由 issue #35 引擎解释）。
|
||||
_FEATURE_NAME_RE = re.compile(r"^[A-Za-z_][A-Za-z0-9_]*$")
|
||||
|
||||
|
||||
@dataclass
|
||||
class ValidationIssue:
|
||||
"""单条校验问题。"""
|
||||
|
||||
code: str # 错误码：missing_field / bad_type / unknown_algorithm / ...
|
||||
path: str # JSON 路径，如 ``features[1].name`` / ``algorithm``
|
||||
message: str # 人类可读描述
|
||||
|
||||
|
||||
@dataclass
|
||||
class ValidationReport:
|
||||
"""校验报告。"""
|
||||
|
||||
issues: List[ValidationIssue] = field(default_factory=list)
|
||||
|
||||
@property
|
||||
def ok(self) -> bool:
|
||||
return not self.issues
|
||||
|
||||
def summary(self) -> str:
|
||||
if self.ok:
|
||||
return "超参包校验通过"
|
||||
by_code: Dict[str, int] = {}
|
||||
for it in self.issues:
|
||||
by_code[it.code] = by_code.get(it.code, 0) + 1
|
||||
parts = [f"{code}×{n}" for code, n in sorted(by_code.items())]
|
||||
return f"超参包校验未通过（{len(self.issues)} 个问题：" + "，".join(parts) + "）"
|
||||
|
||||
|
||||
@dataclass
|
||||
class HyperparamPack:
|
||||
"""解析后的超参包（校验通过的视图）。
|
||||
|
||||
字段对齐 PRD 5.3 超参包示例；可选字段缺失时为 ``None``。
|
||||
"""
|
||||
|
||||
model_id: str
|
||||
template: str
|
||||
algorithm: str
|
||||
features: List[Dict[str, str]]
|
||||
target: str
|
||||
objective: Optional[str] = None
|
||||
hyperparams: Optional[Dict[str, Any]] = None
|
||||
train_window: Optional[str] = None
|
||||
alarm_threshold: Optional[Dict[str, Any]] = None
|
||||
drift_check: Optional[Dict[str, Any]] = None
|
||||
raw: Dict[str, Any] = field(default_factory=dict)
|
||||
|
||||
@property
|
||||
def feature_names(self) -> List[str]:
|
||||
return [f["name"] for f in self.features]
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# 校验主逻辑
|
||||
# ---------------------------------------------------------------------------
|
||||
def validate_pack(pack: Dict[str, Any]) -> ValidationReport:
|
||||
"""校验一个已解析为 dict 的超参包，返回报告（不抛异常）。"""
|
||||
report = ValidationReport()
|
||||
|
||||
def add(code: str, path: str, message: str) -> None:
|
||||
report.issues.append(ValidationIssue(code, path, message))
|
||||
|
||||
# 0. 顶层必须是对象
|
||||
if not isinstance(pack, dict):
|
||||
add("bad_type", "$", f"超参包根节点必须为 JSON 对象，实际为 {type(pack).__name__}")
|
||||
return report
|
||||
|
||||
# 1. 必填字段
|
||||
required: Tuple[str, ...] = ("model_id", "template", "algorithm", "features", "target")
|
||||
for key in required:
|
||||
if key not in pack:
|
||||
add("missing_field", key, f"缺少必填字段：{key}")
|
||||
|
||||
# 2. 标量字段类型与取值
|
||||
model_id = pack.get("model_id")
|
||||
if "model_id" in pack:
|
||||
if not isinstance(model_id, str) or not model_id.strip():
|
||||
add("bad_type", "model_id", "model_id 必须为非空字符串")
|
||||
elif not _FEATURE_NAME_RE.match(model_id):
|
||||
add("bad_format", "model_id", "model_id 含非法字符（仅字母/数字/下划线，首字符非数字）")
|
||||
|
||||
template = pack.get("template")
|
||||
if "template" in pack and (not isinstance(template, str) or not template.strip()):
|
||||
add("bad_type", "template", "template 必须为非空字符串")
|
||||
|
||||
algorithm = pack.get("algorithm")
|
||||
if "algorithm" in pack:
|
||||
if not isinstance(algorithm, str) or not algorithm.strip():
|
||||
add("bad_type", "algorithm", "algorithm 必须为非空字符串")
|
||||
elif algorithm not in KNOWN_ALGORITHMS:
|
||||
known = "、".join(KNOWN_ALGORITHMS)
|
||||
add(
|
||||
"unknown_algorithm",
|
||||
"algorithm",
|
||||
f"未知算法 '{algorithm}'；已知项：{known}（新增结构请走 Model Recipe 插件注册）",
|
||||
)
|
||||
|
||||
target = pack.get("target")
|
||||
if "target" in pack and (not isinstance(target, str) or not target.strip()):
|
||||
add("bad_type", "target", "target 必须为非空字符串")
|
||||
|
||||
# 3. objective（可选）
|
||||
objective = pack.get("objective")
|
||||
if objective is not None:
|
||||
if not isinstance(objective, str) or not objective.strip():
|
||||
add("bad_type", "objective", "objective 若填写必须为非空字符串")
|
||||
elif objective not in KNOWN_OBJECTIVES:
|
||||
known = "、".join(KNOWN_OBJECTIVES)
|
||||
add("unknown_objective", "objective", f"未知目标函数 '{objective}'；已知项：{known}")
|
||||
|
||||
# 4. features（必填，非空列表）
|
||||
features = pack.get("features")
|
||||
if features is None:
|
||||
# missing_field 已在步骤 1 记录
|
||||
pass
|
||||
elif not isinstance(features, list):
|
||||
add("bad_type", "features", f"features 必须为数组，实际为 {type(features).__name__}")
|
||||
elif len(features) == 0:
|
||||
add("empty_features", "features", "features 不能为空（模型至少需要一个输入特征）")
|
||||
else:
|
||||
seen_names: Dict[str, int] = {}
|
||||
for i, feat in enumerate(features):
|
||||
fpath = f"features[{i}]"
|
||||
if not isinstance(feat, dict):
|
||||
add("bad_type", fpath, f"特征项必须为对象，实际为 {type(feat).__name__}")
|
||||
continue
|
||||
name = feat.get("name")
|
||||
spec = feat.get("spec")
|
||||
if not isinstance(name, str) or not name.strip():
|
||||
add("missing_field", f"{fpath}.name", "特征缺少 name 或为空")
|
||||
else:
|
||||
if not _FEATURE_NAME_RE.match(name):
|
||||
add("bad_format", f"{fpath}.name", f"特征名 '{name}' 含非法字符")
|
||||
if name in seen_names:
|
||||
add(
|
||||
"dup_feature",
|
||||
f"{fpath}.name",
|
||||
f"特征名 '{name}' 重复（首次出现在 features[{seen_names[name]}]）",
|
||||
)
|
||||
else:
|
||||
seen_names[name] = i
|
||||
if not isinstance(spec, str) or not spec.strip():
|
||||
add("missing_field", f"{fpath}.spec", f"特征 '{name}' 缺少 spec（FeatureSpec 声明）或为空")
|
||||
|
||||
# 5. hyperparams（可选，对象）
|
||||
hyperparams = pack.get("hyperparams")
|
||||
if hyperparams is not None:
|
||||
if not isinstance(hyperparams, dict):
|
||||
add("bad_type", "hyperparams", f"hyperparams 必须为对象，实际为 {type(hyperparams).__name__}")
|
||||
else:
|
||||
for k, v in hyperparams.items():
|
||||
if not isinstance(k, str) or not k.strip():
|
||||
add("bad_format", "hyperparams", "hyperparams 含空键")
|
||||
|
||||
# 6. train_window（可选，<数><单位>）
|
||||
train_window = pack.get("train_window")
|
||||
if train_window is not None:
|
||||
if not isinstance(train_window, str) or not _TRAIN_WINDOW_RE.match(train_window):
|
||||
add(
|
||||
"bad_format",
|
||||
"train_window",
|
||||
"train_window 必须形如 '<正数><单位>'（单位 d/w/h/m），如 '180d'、'4w'",
|
||||
)
|
||||
|
||||
# 7. alarm_threshold（可选，对象，含 type）
|
||||
alarm = pack.get("alarm_threshold")
|
||||
if alarm is not None:
|
||||
if not isinstance(alarm, dict):
|
||||
add("bad_type", "alarm_threshold", f"alarm_threshold 必须为对象，实际为 {type(alarm).__name__}")
|
||||
else:
|
||||
atype = alarm.get("type")
|
||||
known_alarm_types = ("zscore", "quantile", "absolute")
|
||||
if not isinstance(atype, str) or atype not in known_alarm_types:
|
||||
add(
|
||||
"unknown_alarm_type",
|
||||
"alarm_threshold.type",
|
||||
f"alarm_threshold.type 必须为 {known_alarm_types} 之一",
|
||||
)
|
||||
if atype == "zscore" and "k" not in alarm:
|
||||
add("missing_field", "alarm_threshold.k", "zscore 阈值缺少 k")
|
||||
if atype == "quantile" and "q" not in alarm:
|
||||
add("missing_field", "alarm_threshold.q", "quantile 阈值缺少 q")
|
||||
if atype == "absolute" and "value" not in alarm:
|
||||
add("missing_field", "alarm_threshold.value", "absolute 阈值缺少 value")
|
||||
|
||||
# 8. drift_check（可选，对象，method + limit）
|
||||
drift = pack.get("drift_check")
|
||||
if drift is not None:
|
||||
if not isinstance(drift, dict):
|
||||
add("bad_type", "drift_check", f"drift_check 必须为对象，实际为 {type(drift).__name__}")
|
||||
else:
|
||||
method = drift.get("method")
|
||||
known_methods = ("psi", "ks", "chi2")
|
||||
if not isinstance(method, str) or method not in known_methods:
|
||||
add(
|
||||
"unknown_drift_method",
|
||||
"drift_check.method",
|
||||
f"drift_check.method 必须为 {known_methods} 之一",
|
||||
)
|
||||
limit = drift.get("limit")
|
||||
if limit is None:
|
||||
add("missing_field", "drift_check.limit", "drift_check 缺少 limit")
|
||||
elif not isinstance(limit, (int, float)) or isinstance(limit, bool):
|
||||
add("bad_type", "drift_check.limit", "drift_check.limit 必须为数值")
|
||||
elif not (0 < limit <= 1):
|
||||
add("bad_range", "drift_check.limit", "drift_check.limit 必须在 (0, 1] 范围内")
|
||||
|
||||
return report
|
||||
|
||||
|
||||
def load_pack(pack: Dict[str, Any]) -> HyperparamPack:
|
||||
"""校验并把 dict 装配为 :class:`HyperparamPack`；校验失败抛 ``ValueError``。
|
||||
|
||||
供训练 / 推理流水线在加载超参包时 fail-fast 使用。
|
||||
"""
|
||||
report = validate_pack(pack)
|
||||
if not report.ok:
|
||||
raise ValueError(report.summary())
|
||||
|
||||
return HyperparamPack(
|
||||
model_id=pack["model_id"],
|
||||
template=pack["template"],
|
||||
algorithm=pack["algorithm"],
|
||||
features=list(pack["features"]),
|
||||
target=pack["target"],
|
||||
objective=pack.get("objective"),
|
||||
hyperparams=pack.get("hyperparams"),
|
||||
train_window=pack.get("train_window"),
|
||||
alarm_threshold=pack.get("alarm_threshold"),
|
||||
drift_check=pack.get("drift_check"),
|
||||
raw=dict(pack),
|
||||
)
|
||||
|
||||
|
||||
def validate_pack_file(path: str) -> ValidationReport:
|
||||
"""读取 JSON 文件并校验；文件 / JSON 解析错误也记入报告（不抛异常）。"""
|
||||
report = ValidationReport()
|
||||
|
||||
def add(code: str, msg: str) -> None:
|
||||
report.issues.append(ValidationIssue(code, "$", msg))
|
||||
|
||||
if not os.path.exists(path):
|
||||
add("file_not_found", f"超参包文件不存在：{path}")
|
||||
return report
|
||||
try:
|
||||
with open(path, "r", encoding="utf-8") as fh:
|
||||
text = fh.read()
|
||||
except OSError as exc:
|
||||
add("file_read_error", f"读取超参包失败：{exc}")
|
||||
return report
|
||||
try:
|
||||
pack = json.loads(text)
|
||||
except json.JSONDecodeError as exc:
|
||||
add("json_parse_error", f"超参包不是合法 JSON：{exc}")
|
||||
return report
|
||||
|
||||
inner = validate_pack(pack)
|
||||
report.issues.extend(inner.issues)
|
||||
return report
|
||||
@@ -0,0 +1,16 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""测试引导：把 `core/model-framework` 以包名 `model_framework` 挂载到 sys.modules。
|
||||
|
||||
目录名 `model-framework` 含连字符，无法直接以包名 import；挂载后模块内相对导入
|
||||
（`from .hyperparam import ...`）在 unittest 发现机制下可正常解析。
|
||||
"""
|
||||
import os
|
||||
import sys
|
||||
import types
|
||||
|
||||
MF_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
||||
sys.path.insert(0, MF_DIR)
|
||||
if "model_framework" not in sys.modules:
|
||||
pkg = types.ModuleType("model_framework")
|
||||
pkg.__path__ = [MF_DIR]
|
||||
sys.modules["model_framework"] = pkg
|
||||
@@ -0,0 +1,239 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""超参包 JSON Schema 校验器测试（issue #39）。
|
||||
|
||||
覆盖：合法包通过 / 必填缺失 / 未知算法 / 特征重复与缺 spec / 目标函数 /
|
||||
train_window 格式 / alarm_threshold / drift_check / 文件加载与 JSON 解析错误。
|
||||
"""
|
||||
import os
|
||||
import tempfile
|
||||
import unittest
|
||||
|
||||
import _bootstrap # noqa: F401 挂载包名
|
||||
|
||||
from model_framework.hyperparam import (
|
||||
HyperparamPack,
|
||||
load_pack,
|
||||
validate_pack,
|
||||
validate_pack_file,
|
||||
)
|
||||
|
||||
|
||||
def _good_pack():
|
||||
"""PRD 5.3 示例的合法超参包。"""
|
||||
return {
|
||||
"model_id": "quality_predict_ti",
|
||||
"template": "iAOP-Template-Ti",
|
||||
"algorithm": "xgboost",
|
||||
"features": [
|
||||
{"name": "EMA_CLF_TEMP_5m", "spec": "EMA(CLF-01.TEMP, 5m)"},
|
||||
{"name": "RollingStd_CL2_10", "spec": "RollingStd(CLF-01.CL2, 10)"},
|
||||
],
|
||||
"target": "Ti_purity",
|
||||
"objective": "reg:squarederror",
|
||||
"hyperparams": {"max_depth": 6, "eta": 0.1, "n_estimators": 300},
|
||||
"train_window": "180d",
|
||||
"alarm_threshold": {"type": "zscore", "k": 3.0},
|
||||
"drift_check": {"method": "psi", "limit": 0.2},
|
||||
}
|
||||
|
||||
|
||||
class ValidatePackTest(unittest.TestCase):
|
||||
def test_good_pack_passes(self):
|
||||
report = validate_pack(_good_pack())
|
||||
self.assertTrue(report.ok, report.summary())
|
||||
self.assertEqual(report.issues, [])
|
||||
|
||||
def test_minimal_pack_passes(self):
|
||||
report = validate_pack(
|
||||
{
|
||||
"model_id": "anomaly_resin",
|
||||
"template": "iAOP-Template-Resin",
|
||||
"algorithm": "isolation_forest",
|
||||
"features": [{"name": "F1", "spec": "RateOfChange(炉压)"}],
|
||||
"target": "anomaly_score",
|
||||
}
|
||||
)
|
||||
self.assertTrue(report.ok, report.summary())
|
||||
|
||||
def test_missing_required_field(self):
|
||||
pack = _good_pack()
|
||||
del pack["algorithm"]
|
||||
report = validate_pack(pack)
|
||||
self.assertFalse(report.ok)
|
||||
codes = [i.code for i in report.issues]
|
||||
self.assertIn("missing_field", codes)
|
||||
self.assertTrue(any(i.path == "algorithm" for i in report.issues))
|
||||
|
||||
def test_missing_all_required(self):
|
||||
report = validate_pack({})
|
||||
self.assertFalse(report.ok)
|
||||
codes = [i.code for i in report.issues]
|
||||
# 5 个必填字段全部缺失
|
||||
self.assertEqual(codes.count("missing_field"), 5)
|
||||
|
||||
def test_unknown_algorithm_lists_known(self):
|
||||
pack = _good_pack()
|
||||
pack["algorithm"] = "magic_boost"
|
||||
report = validate_pack(pack)
|
||||
self.assertFalse(report.ok)
|
||||
issue = next(i for i in report.issues if i.code == "unknown_algorithm")
|
||||
self.assertIn("xgboost", issue.message)
|
||||
self.assertIn("magic_boost", issue.message)
|
||||
|
||||
def test_empty_features_rejected(self):
|
||||
pack = _good_pack()
|
||||
pack["features"] = []
|
||||
report = validate_pack(pack)
|
||||
codes = [i.code for i in report.issues]
|
||||
self.assertIn("empty_features", codes)
|
||||
|
||||
def test_duplicate_feature_name(self):
|
||||
pack = _good_pack()
|
||||
pack["features"] = [
|
||||
{"name": "F1", "spec": "EMA(A, 5m)"},
|
||||
{"name": "F1", "spec": "EMA(B, 5m)"},
|
||||
]
|
||||
report = validate_pack(pack)
|
||||
codes = [i.code for i in report.issues]
|
||||
self.assertIn("dup_feature", codes)
|
||||
|
||||
def test_feature_missing_spec(self):
|
||||
pack = _good_pack()
|
||||
pack["features"] = [{"name": "F1", "spec": ""}]
|
||||
report = validate_pack(pack)
|
||||
codes = [i.code for i in report.issues]
|
||||
self.assertIn("missing_field", codes)
|
||||
|
||||
def test_feature_bad_name_format(self):
|
||||
pack = _good_pack()
|
||||
pack["features"] = [{"name": "1bad", "spec": "EMA(A,5m)"}]
|
||||
report = validate_pack(pack)
|
||||
codes = [i.code for i in report.issues]
|
||||
self.assertIn("bad_format", codes)
|
||||
|
||||
def test_unknown_objective(self):
|
||||
pack = _good_pack()
|
||||
pack["objective"] = "reg:magic"
|
||||
report = validate_pack(pack)
|
||||
self.assertTrue(any(i.code == "unknown_objective" for i in report.issues))
|
||||
|
||||
def test_bad_train_window_format(self):
|
||||
for bad in ["180", "180days", "abc", "", "1y"]:
|
||||
pack = _good_pack()
|
||||
pack["train_window"] = bad
|
||||
report = validate_pack(pack)
|
||||
self.assertTrue(
|
||||
any(i.code == "bad_format" and i.path == "train_window" for i in report.issues),
|
||||
f"应拒绝非法 train_window: {bad!r}",
|
||||
)
|
||||
|
||||
def test_good_train_window_units(self):
|
||||
for good in ["180d", "4w", "72h", "30m"]:
|
||||
pack = _good_pack()
|
||||
pack["train_window"] = good
|
||||
report = validate_pack(pack)
|
||||
self.assertTrue(report.ok, f"应接受合法 train_window: {good!r} -> {report.summary()}")
|
||||
|
||||
def test_alarm_threshold_zscore_missing_k(self):
|
||||
pack = _good_pack()
|
||||
pack["alarm_threshold"] = {"type": "zscore"}
|
||||
report = validate_pack(pack)
|
||||
self.assertTrue(any(i.code == "missing_field" and i.path == "alarm_threshold.k" for i in report.issues))
|
||||
|
||||
def test_alarm_threshold_unknown_type(self):
|
||||
pack = _good_pack()
|
||||
pack["alarm_threshold"] = {"type": "voodoo"}
|
||||
report = validate_pack(pack)
|
||||
self.assertTrue(any(i.code == "unknown_alarm_type" for i in report.issues))
|
||||
|
||||
def test_drift_check_limit_range(self):
|
||||
for bad_limit in [0, 1.5, -0.1]:
|
||||
pack = _good_pack()
|
||||
pack["drift_check"] = {"method": "psi", "limit": bad_limit}
|
||||
report = validate_pack(pack)
|
||||
self.assertTrue(
|
||||
any(i.code == "bad_range" for i in report.issues),
|
||||
f"应拒绝非法 drift_check.limit: {bad_limit}",
|
||||
)
|
||||
|
||||
def test_drift_check_unknown_method(self):
|
||||
pack = _good_pack()
|
||||
pack["drift_check"] = {"method": "magic", "limit": 0.2}
|
||||
report = validate_pack(pack)
|
||||
self.assertTrue(any(i.code == "unknown_drift_method" for i in report.issues))
|
||||
|
||||
def test_root_not_object(self):
|
||||
report = validate_pack([1, 2, 3]) # type: ignore[arg-type]
|
||||
self.assertFalse(report.ok)
|
||||
self.assertEqual(report.issues[0].code, "bad_type")
|
||||
|
||||
def test_report_summary_ok(self):
|
||||
self.assertEqual(validate_pack(_good_pack()).summary(), "超参包校验通过")
|
||||
|
||||
def test_report_summary_aggregates(self):
|
||||
pack = _good_pack()
|
||||
del pack["algorithm"]
|
||||
del pack["target"]
|
||||
summary = validate_pack(pack).summary()
|
||||
self.assertIn("2 个问题", summary)
|
||||
self.assertIn("missing_field", summary)
|
||||
|
||||
|
||||
class LoadPackTest(unittest.TestCase):
|
||||
def test_load_good_returns_pack(self):
|
||||
pack = load_pack(_good_pack())
|
||||
self.assertIsInstance(pack, HyperparamPack)
|
||||
self.assertEqual(pack.model_id, "quality_predict_ti")
|
||||
self.assertEqual(pack.algorithm, "xgboost")
|
||||
self.assertEqual(pack.feature_names, ["EMA_CLF_TEMP_5m", "RollingStd_CL2_10"])
|
||||
self.assertEqual(pack.hyperparams["max_depth"], 6)
|
||||
|
||||
def test_load_bad_raises(self):
|
||||
pack = _good_pack()
|
||||
del pack["features"]
|
||||
with self.assertRaises(ValueError) as cm:
|
||||
load_pack(pack)
|
||||
self.assertIn("超参包校验未通过", str(cm.exception))
|
||||
|
||||
|
||||
class ValidatePackFileTest(unittest.TestCase):
|
||||
def setUp(self):
|
||||
self._tmp = tempfile.mkdtemp()
|
||||
self.path = os.path.join(self._tmp, "pack.json")
|
||||
|
||||
def _write(self, text):
|
||||
with open(self.path, "w", encoding="utf-8") as fh:
|
||||
fh.write(text)
|
||||
return self.path
|
||||
|
||||
def test_valid_file_passes(self):
|
||||
import json
|
||||
|
||||
self._write(json.dumps(_good_pack()))
|
||||
report = validate_pack_file(self.path)
|
||||
self.assertTrue(report.ok, report.summary())
|
||||
|
||||
def test_missing_file(self):
|
||||
report = validate_pack_file(os.path.join(self._tmp, "nope.json"))
|
||||
self.assertFalse(report.ok)
|
||||
self.assertEqual(report.issues[0].code, "file_not_found")
|
||||
|
||||
def test_bad_json(self):
|
||||
self._write("{ not json ")
|
||||
report = validate_pack_file(self.path)
|
||||
self.assertFalse(report.ok)
|
||||
self.assertEqual(report.issues[0].code, "json_parse_error")
|
||||
|
||||
def test_file_with_schema_errors(self):
|
||||
import json
|
||||
|
||||
bad = _good_pack()
|
||||
bad["algorithm"] = "unknown_thing"
|
||||
self._write(json.dumps(bad))
|
||||
report = validate_pack_file(self.path)
|
||||
self.assertFalse(report.ok)
|
||||
self.assertEqual(report.issues[0].code, "unknown_algorithm")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user