diff --git a/backend/app/engine/evidence.py b/backend/app/engine/evidence.py index 41496690..7c74a1b7 100644 --- a/backend/app/engine/evidence.py +++ b/backend/app/engine/evidence.py @@ -4751,7 +4751,11 @@ _DECOMPOSE_NUMBERS = ("base", "target", "delta", "pct", "share", "contrib", "fla def _decompose_row(result: dict[str, Any]) -> tuple[str, list[Any]] | None: """变化拆解的结果表里判定点名的那一段:(这一段的叫法, 那一行)。段按维度列(params.by;乘积模式是 factor 一列)对 - verdict.subjects 的第一项,对不上就取第一行(结果表按变化的绝对值从大到小排)。叫法是维度值用「·」连起来。""" + verdict.subjects 的第一项,对不上就取第一行(结果表按变化的绝对值从大到小排)。维度值和叫法都按算子写段的同一个函数 + (app.analysis.phrases.key_text:空值写「(空)」、布尔写「是 / 否」,两列用「·」连):subjects 就是这么写的,拿原值的 + str 去比,空值、布尔维度的段对不上,代入式会退回第一行。""" + from app.analysis.phrases import key_text + table = result.get("table") if isinstance(result.get("table"), dict) else {} rows = [r for r in table.get("rows") or [] if isinstance(r, list)] columns = [str(c) for c in table.get("columns") or []] @@ -4768,8 +4772,8 @@ def _decompose_row(result: dict[str, Any]) -> tuple[str, list[Any]] | None: picked = rows[0] if subjects and at and isinstance(subjects[0], list): wanted = [str(v) for v in subjects[0]] - picked = next((row for row in rows if [str(row[i]) for i in at if i < len(row)] == wanted), rows[0]) - name = "·".join(str(picked[i]) for i in at if i < len(picked) and picked[i] is not None) + picked = next((row for row in rows if [key_text([row[i]]) for i in at if i < len(row)] == wanted), rows[0]) + name = key_text([picked[i] for i in at if i < len(picked)]) if at else "" return name or "这一段", picked diff --git a/backend/tests/test_analysis_evidence.py b/backend/tests/test_analysis_evidence.py index 84c922de..4bc1449f 100644 --- a/backend/tests/test_analysis_evidence.py +++ b/backend/tests/test_analysis_evidence.py @@ -601,6 +601,25 @@ def test_formulas_substitute_the_key_numbers(): assert analysis_formulas(fake_result(values={})) == [] +@pytest.mark.parametrize("keys, named", [((None, "华南", "华北"), "(空)"), ((True, False, None), "是")]) +def test_the_formula_names_the_segment_the_verdict_names(keys, named): + """代入式里点名的那一段按判定的 subjects 找,维度值按算子写段的同一个函数比(空值写「(空)」、布尔写「是 / 否」)。 + 这里判定点名的段不在结果表第一行(第一行是反方向变化最大的那段):以前拿原值的 str 比,对不上就退回第一行,代入式写成了 + 第一行那段。""" + store = Store() + driver, opposite, other = keys # 变化:driver -220、opposite +250、other -180,总共 -150 + snap = {"columns": ["region", "week", "gmv"], "column_types": {"region": "text", "week": "text", "gmv": "number"}, + "rows": [[driver, "W41", 1000], [driver, "W42", 780], [opposite, "W41", 500], [opposite, "W42", 750], + [other, "W41", 600], [other, "W42", 420]], + "sql": "SELECT region, week, SUM(gmv) AS gmv FROM orders GROUP BY region, week", "source": "shop", + "truncated": False} + result = real_result(store.put(snap), snap) + assert result["verdict"]["code"] == "driver" and result["verdict"]["subjects"] == [[named]] + assert [r[0] for r in result["table"]["rows"]] == [opposite, driver, other] # 按变化的绝对值排 + texts = {f["label"]: f["text"] for f in analysis_formulas(result)} + assert texts["占比"] == f"{named}的占比 = {named}的变化 ÷ 总变化 = -220 ÷ (-150) = 146.7%" + + def test_formulas_for_factor_ratio_and_baseline(): store = Store() factor_snap = {"columns": ["week", "gmv", "orders", "aov"],