| 1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889 |
- """V3-M3A: platform heat 归一化单测;R3(2026-06-12)改配置驱动复合后扩充。"""
- from __future__ import annotations
- from math import log10
- from content_agent.business_modules.content_discovery.platform_heat import heat_score
- def _stat(digg, **extra):
- return {"digg_count": digg, **extra}
- # --- R3 第一步:点赞单信号,与 R3 前行为字节等价(锚点现读自 profile heat 段) ---
- def test_heat_floor_maps_to_zero():
- assert heat_score(_stat(10000), "douyin") == 0.0
- assert heat_score(_stat(3), "douyin") == 0.0
- def test_heat_ceil_maps_to_one():
- assert heat_score(_stat(1000000), "douyin") == 1.0
- assert heat_score(_stat(5034215), "douyin") == 1.0
- def test_heat_midpoint_between_zero_and_one():
- # geometric midpoint of (1e4, 1e6) is 1e5 -> ~0.5
- assert abs(heat_score(_stat(100000), "douyin") - 0.5) < 0.01
- def test_heat_missing_or_zero_digg_is_zero():
- assert heat_score(_stat(None), "douyin") == 0.0
- assert heat_score(_stat(0), "shipinhao") == 0.0
- assert heat_score(_stat("not-a-number"), "douyin") == 0.0
- def test_heat_no_heat_block_platform_uses_default_anchor():
- # 无 heat 段的平台(github profile 无 heat 段)回退默认锚点 (100, 1e5):
- # below floor -> 0, at/above ceil -> 1。
- assert heat_score(_stat(50), "github") == 0.0
- assert heat_score(_stat(100000), "github") == 1.0
- def test_heat_shipinhao_low_digg_not_unfairly_zero():
- # 92 likes is meaningful on shipinhao (anchors 50..5e4) though tiny on douyin.
- assert heat_score(_stat(92), "shipinhao") > 0.05
- assert heat_score(_stat(92), "douyin") == 0.0
- # --- R3 复合机制守卫(放宽到多字段后这些仍成立,无需改) ---
- def test_heat_single_signal_equals_legacy_digg_only():
- # 单信号配置 == 老的纯点赞归一化(机制中性证明)。
- digg = 200000
- expected = round((log10(digg + 1) - log10(10000)) / (log10(1000000) - log10(10000)), 4)
- assert heat_score(_stat(digg), "douyin") == expected
- def test_heat_ignores_zero_and_missing_fields_in_composite(monkeypatch):
- # 复合时 0 值字段被排除、权重在在场字段间重新归一。
- from content_agent.business_modules.content_discovery import platform_heat
- def fake_signals(platform):
- return [
- {"field": "digg_count", "weight": 0.5, "floor": 100, "ceil": 100000},
- {"field": "comment_count", "weight": 0.5, "floor": 10, "ceil": 10000},
- ]
- monkeypatch.setattr(platform_heat, "_heat_signals", fake_signals)
- # comment 为 0 → 只剩 digg,权重重新归一为 1.0 → 等于纯 digg 归一
- digg_only = heat_score({"digg_count": 10000, "comment_count": 0}, "x")
- expected = round((log10(10001) - log10(100)) / (log10(100000) - log10(100)), 4)
- assert digg_only == expected
- def test_heat_weighted_average_of_two_signals(monkeypatch):
- from content_agent.business_modules.content_discovery import platform_heat
- def fake_signals(platform):
- return [
- {"field": "digg_count", "weight": 0.5, "floor": 100, "ceil": 100000},
- {"field": "comment_count", "weight": 0.5, "floor": 10, "ceil": 10000},
- ]
- monkeypatch.setattr(platform_heat, "_heat_signals", fake_signals)
- score = heat_score({"digg_count": 10000, "comment_count": 1000}, "x")
- d = (log10(10001) - log10(100)) / (log10(100000) - log10(100))
- c = (log10(1001) - log10(10)) / (log10(10000) - log10(10))
- assert score == round((0.5 * d + 0.5 * c) / 1.0, 4)
|
