mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-20 02:01:58 +00:00
分析结论记录说法模板版本:复算时版本不同不误报文字差异
This commit is contained in:
@@ -91,6 +91,7 @@ def build(env: Env, params: dict[str, Any], used: Sequence[str], *, status: str
|
||||
"verdict": dict(verdict) if verdict is not None else None, # type: ignore[typeddict-item]
|
||||
"criterion": criterion,
|
||||
"phrase": phrase,
|
||||
"phrase_version": ph.PHRASE_VERSION,
|
||||
"caveats": [dict(c) for c in caveats], # type: ignore[misc]
|
||||
"hypothesis": env.hypothesis,
|
||||
}
|
||||
|
||||
@@ -312,6 +312,9 @@ class AnalysisResult(TypedDict, total=False):
|
||||
verdict: Verdict | None
|
||||
criterion: str # 判定标准,一句,照实际阈值写
|
||||
phrase: str # 判定语(app/analysis/phrases 渲染),[[a:A3]] 渲染成它
|
||||
#: 渲染判定语时用的模板版本(app/analysis/phrases.PHRASE_VERSION):版本不同时 phrase、caveats 不参与复算比较,
|
||||
#: 旧结论没有这一项按没有记录版本算
|
||||
phrase_version: int
|
||||
caveats: list[Caveat]
|
||||
hypothesis: str | None
|
||||
|
||||
@@ -436,8 +439,12 @@ FORMAT_CONVS: tuple[str, ...] = (CONV_PCT, CONV_POINTS)
|
||||
|
||||
|
||||
def same(a: Mapping[str, Any], b: Mapping[str, Any]) -> bool:
|
||||
"""两份结论(封存的、复算的)是否逐字段相同(hypothesis 除外)。"""
|
||||
"""两份结论(封存的、复算的)是否逐字段相同(hypothesis 除外)。说法模板版本不同时,渲染出来的 phrase、caveats
|
||||
和版本号本身都不比:改一句模板不该让旧结论报「复算不一致」;版本相同(两边都没有版本也算相同)仍严格比 phrase、
|
||||
caveats。"""
|
||||
keys = (set(a) | set(b)) - NOT_RECOMPUTED
|
||||
if a.get("phrase_version") != b.get("phrase_version"):
|
||||
keys -= {"phrase", "caveats", "phrase_version"}
|
||||
return all(a.get(k) == b.get(k) for k in keys)
|
||||
|
||||
|
||||
|
||||
@@ -19,6 +19,10 @@
|
||||
|
||||
**附带说明**:判定语后面用「;」接,最多两条(按 PHRASE_ORDER 的先后),其余只在证据面板。度量类型没标、以各段之和为
|
||||
总数这类方法上的交代不进判定语。
|
||||
|
||||
**版本**(PHRASE_VERSION):判定语是结论的一部分,复算时逐字段比。改了判定语模板、数的写法、附带说明的拼接规则就升
|
||||
版本,结论里记下渲染时用的那一版;复算遇到版本不同时渲染文字不参与比较,免得改一句模板就让旧结论误报「复算不一致」。
|
||||
engine/analysis_claims 的说法表词表改了也要升:label 消毒按它认判断词,消完的叫法会变。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
@@ -31,6 +35,8 @@ from typing import Any
|
||||
from app.analysis.contract import OP_LABELS
|
||||
from app.analysis.numbers import fixed, to_fraction
|
||||
|
||||
#: 判定语模板的版本(见文件头):改了模板、数的写法、附带说明的拼接规则,或者 engine/analysis_claims 的说法表词表就升
|
||||
PHRASE_VERSION = 1
|
||||
#: 附带说明进判定语的先后;不在这里的(measure_unlabeled、headline_basis、partial_periods)只在证据面板
|
||||
PHRASE_ORDER = ("comparability", "untested", "simpson", "offset", "outlier_sensitive", "small_sample", "record_high",
|
||||
"record_low", "multiple_comparisons", "weekday_effect", "small_denominator", "null_values")
|
||||
@@ -572,7 +578,7 @@ CRITERION_COMPARABLE = ("只差日期、窗口一样长、是整期、星期构
|
||||
|
||||
__all__ = [
|
||||
"CRITERION_COMPARABLE", "EMPTY_KEY", "HEADLINE_BASIS", "KEY_JOIN", "LABEL_MAX", "MEASURE_WORDS", "NO_DATES",
|
||||
"PHRASE_CAVEATS", "PHRASE_ORDER", "UNKNOWN_WEEKS", "UNVERIFIED_COMPLETE", "WEEKDAY_EFFECT", "amount", "band_text",
|
||||
"PHRASE_CAVEATS", "PHRASE_ORDER", "PHRASE_VERSION", "UNKNOWN_WEEKS", "UNVERIFIED_COMPLETE", "WEEKDAY_EFFECT", "amount", "band_text",
|
||||
"baseline_flat_phrase", "baseline_phrase", "caveat", "change", "clean_label", "comparable_phrase",
|
||||
"criterion_additive", "criterion_baseline", "criterion_differ", "criterion_factor", "criterion_ratio",
|
||||
"criterion_trend", "current_text", "decompose_concentrated", "decompose_dispersed", "decompose_driver",
|
||||
|
||||
@@ -4797,7 +4797,7 @@ _RESULT_FIELDS = {
|
||||
"op": "算子", "mode": "模式", "method": "方法版本", "params": "参数", "settings": "阈值", "inputs": "输入",
|
||||
"probes": "现查结果", "status": "算成没算成", "reason_code": "算不了的原因", "reason": "算不了的原因",
|
||||
"table": "结果表", "values": "具名的数", "verdict": "判定", "criterion": "判定标准", "phrase": "判定语",
|
||||
"caveats": "附带说明",
|
||||
"phrase_version": "说法模板版本", "caveats": "附带说明",
|
||||
}
|
||||
#: app.analysis.recompute 取不回输入(CallError 的 code input_missing)时那句以「复算」开头(复算取不回 Q2 的查询快照):
|
||||
#: 接在「复算失败:」后面时去掉,不说两遍
|
||||
@@ -4824,32 +4824,50 @@ def analysis_recheck(result: dict[str, Any], load: Callable[[str], Any], *,
|
||||
|
||||
方法升过版(结论里记的 method 和现在的 contract.METHODS[(op, mode)] 不同)时 method_now 是现在的那一版:复算成立的照实写
|
||||
复算结果同不同(方法版本那一项本来就不同,不算进去),现在的方法按封存的输入不成立(recompute 抛 CallError)的写明原因。
|
||||
说法模板升过版(结论里记的 phrase_version 和现在的 phrases.PHRASE_VERSION 不同,旧结论没有这一项也算)时 phrase_now 是
|
||||
现在的那一版:渲染出来的 phrase、caveats 不参与比较,也不列成差异,文字里写明升过版。
|
||||
|
||||
复算出错(输入取不回、调用的写法现在不认、出了意外)不往外抛,照实写「复算失败:原因」。"""
|
||||
from app import analysis
|
||||
from app.analysis import phrases as ph
|
||||
from app.analysis.contract import METHODS, NOT_RECOMPUTED, CallError, same
|
||||
from app.core.errors import describe_exception
|
||||
|
||||
method = str(result.get("method") or "")
|
||||
now = METHODS.get((result.get("op"), result.get("mode")))
|
||||
upgraded = isinstance(now, str) and now != method
|
||||
out: dict[str, Any] = {"method": method, **({"method_now": now} if upgraded else {})}
|
||||
phrase = result.get("phrase_version")
|
||||
phrase_now = ph.PHRASE_VERSION
|
||||
phrase_upgraded = phrase != phrase_now
|
||||
out: dict[str, Any] = {"method": method, **({"method_now": now} if upgraded else {}),
|
||||
**({"phrase_version": phrase, "phrase_now": phrase_now} if phrase_upgraded else {})}
|
||||
head = f"运行时用的是 {method},现在是 {now}" if upgraded else ""
|
||||
phrase_head = f"说法模板没有记录版本,现在是 v{phrase_now}" if phrase is None \
|
||||
else f"说法模板升过版(v{phrase} → v{phrase_now})"
|
||||
lead = ";".join(x for x in (head, phrase_head if phrase_upgraded else "") if x)
|
||||
try:
|
||||
again = analysis.recompute(result, load, load_result=load_result) # type: ignore[arg-type]
|
||||
except CallError as e:
|
||||
why = _recheck_reason(e)
|
||||
text = f"{head}:现在的方法按封存的输入复算不了({why})" if upgraded else f"复算失败:{why}"
|
||||
if upgraded:
|
||||
text = f"{lead}:现在的方法按封存的输入复算不了({why})"
|
||||
else:
|
||||
text = f"复算失败:{why}" + (f";{phrase_head}" if phrase_upgraded else "")
|
||||
return {"status": "failed", **out, "text": text}
|
||||
except Exception as e: # noqa: BLE001 - 复算出了意外:照实说复算不了,不能让证据面板、导出跟着出错
|
||||
return {"status": "failed", **out, "text": f"复算失败:{describe_exception(e)}"}
|
||||
return {"status": "failed", **out,
|
||||
"text": f"复算失败:{describe_exception(e)}" + (f";{phrase_head}" if phrase_upgraded else "")}
|
||||
if not isinstance(again, dict):
|
||||
return {"status": "failed", **out, "text": "复算失败:复算没有交回结论"}
|
||||
return {"status": "failed", **out,
|
||||
"text": "复算失败:复算没有交回结论" + (f";{phrase_head}" if phrase_upgraded else "")}
|
||||
ok = same({**result, "method": now} if upgraded else result, again)
|
||||
skip = NOT_RECOMPUTED | ({"method"} if upgraded else set())
|
||||
if phrase_upgraded:
|
||||
skip |= {"phrase", "caveats", "phrase_version"}
|
||||
fields = sorted(k for k in set(result) | set(again) if k not in skip and result.get(k) != again.get(k))
|
||||
out["status"] = "same" if ok else "different"
|
||||
if upgraded:
|
||||
out["text"] = f"{head},复算结果{'相同' if ok else '不同'}"
|
||||
if lead:
|
||||
out["text"] = f"{lead},复算结果{'相同' if ok else '不同'}"
|
||||
elif ok:
|
||||
out["text"] = "复算一致:按封存的输入重新计算,结论逐项相同"
|
||||
else:
|
||||
|
||||
@@ -48,6 +48,17 @@ def test_recompute_comparison_ignores_only_the_hypothesis():
|
||||
assert not c.same(sealed, {**sealed, "phrase": "改过的判定语"})
|
||||
|
||||
|
||||
def test_recompute_comparison_ignores_the_rendered_phrase_when_versions_differ():
|
||||
"""说法模板升过版:旧结论记的 phrase、caveats 不参与比较;版本相同(包括两边都没有版本)仍严格比。"""
|
||||
sealed = {**fake_result(), "phrase_version": 1}
|
||||
upgraded = {**sealed, "phrase_version": 2, "phrase": "新版判定语",
|
||||
"caveats": [{"code": "offset", "text": "新版附带说明"}]}
|
||||
assert c.same(sealed, upgraded)
|
||||
assert not c.same(sealed, {**sealed, "phrase": "改过的判定语"})
|
||||
assert not c.same({k: v for k, v in sealed.items() if k != "phrase_version"},
|
||||
{k: v for k, v in sealed.items() if k != "phrase_version"} | {"phrase": "改过的判定语"})
|
||||
|
||||
|
||||
def test_every_op_has_a_method_and_a_tool_name():
|
||||
for op in c.OPS:
|
||||
modes = c.MODES.get(op, (None,))
|
||||
|
||||
@@ -612,6 +612,22 @@ def test_recheck_says_what_differs_and_what_is_missing():
|
||||
assert analysis_recheck(sealed, lambda sha: other)["status"] == "different"
|
||||
|
||||
|
||||
def test_recheck_says_the_phrase_template_was_upgraded(monkeypatch):
|
||||
"""说法模板升过版:复算按新模板渲染,旧结论改过的 phrase 不算差异,文字写明升过版;旧结论没记版本也照实说。"""
|
||||
store = Store()
|
||||
q = store.put(SNAPSHOT)
|
||||
sealed = {**real_result(q), "phrase_version": 1, "phrase": "旧模板的判定语"}
|
||||
monkeypatch.setattr("app.analysis.phrases.PHRASE_VERSION", 2)
|
||||
out = analysis_recheck(sealed, store.get)
|
||||
assert out == {"status": "same", "method": "decompose.additive@1", "phrase_version": 1, "phrase_now": 2,
|
||||
"text": "说法模板升过版(v1 → v2),复算结果相同"}
|
||||
assert "fields" not in out
|
||||
old = {k: v for k, v in sealed.items() if k != "phrase_version"}
|
||||
out = analysis_recheck(old, store.get)
|
||||
assert out["status"] == "same" and out["phrase_version"] is None and out["phrase_now"] == 2
|
||||
assert out["text"] == "说法模板没有记录版本,现在是 v2,复算结果相同"
|
||||
|
||||
|
||||
def test_recheck_names_both_versions_when_the_method_was_upgraded(monkeypatch):
|
||||
store = Store()
|
||||
q = store.put(SNAPSHOT)
|
||||
|
||||
Reference in New Issue
Block a user