Light1YT
81df075ccc
Add demand_normalization.py: norm = clamp(exp(beta·(rate_future − rate_window_avg)), 0.5, 1.2). Discounts a sales pace observed under one rate regime when projecting into a different (higher-rate) future — ТЗ §9.4: don't carry a low-rate boom pace forward into high rates. - beta reused from §9.6 rate_sensitivity (PR3, shrunk slope on Δln, gated <0); rate_window_avg = mean key_rate over the §9.6 window (PR2 macro_series). - Honesty gate: β None / §9.6 confidence='low' / no rate window → norm=1.0, applied=False, confidence='low' (no naive boom, no fabricated discount). applied=False is distinct from a trustworthy coefficient==1.0 (regimes match). confidence never exceeds §9.6's. - Pure normalization_factor() + frozen DemandNormalization + as_dict(). - 26 unit tests, no live DB (mocked PR2/PR3). ADVISORY: not wired into endpoints. Completes the deterministic §9.4/§9.5/§9.6 forecasting engine (PR2/PR1/PR3/PR4/PR5). PR6 follow-up: OverflowError guard on exported normalization_factor (unreachable via orchestrator today, β bounded).
92 lines
3 KiB
Python
92 lines
3 KiB
Python
"""Forecasting services — детерминированный форсайт-слой Site Finder v2.
|
|
|
|
#951 (Site Finder v2 / «GG-форсайт», EPIC 7 «Чувствительность к ключевой ставке»).
|
|
|
|
Этот пакет — фундамент data-independent логики прогноза: monthly макро-ряды,
|
|
классификатор режима ставки, лаговые помощники. Всё ДЕТЕРМИНИРОВАННО, БЕЗ LLM.
|
|
|
|
Слои (по PR):
|
|
• macro_series (#951b) — monthly макро-ряд + классификатор режима ставки (X-ось §9.6).
|
|
• sales_series (#951c) — monthly ряд продаж по сегменту (Y-ось §9.6).
|
|
• rate_sensitivity (#951d) — §9.6 чувствительность продаж к key_rate (CORE, ADVISORY).
|
|
• macro_coefficient (#951e) — §9.5 макро-коэффициент (композитный множитель, ADVISORY).
|
|
• demand_normalization (#951f) — §9.4 нормализация спроса под смену режима ставки (ADVISORY).
|
|
|
|
Источники данных:
|
|
• макро — таблица macro_indicator через reader site_finder/macro.py (reuse).
|
|
• продажи — objective_corpus_room_month / objective_lots (см. sales_series).
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from app.services.forecasting.demand_normalization import (
|
|
DemandNormalization,
|
|
compute_demand_normalization,
|
|
normalization_factor,
|
|
)
|
|
from app.services.forecasting.macro_coefficient import (
|
|
MacroCoefficient,
|
|
assemble_coefficient,
|
|
compute_macro_coefficient,
|
|
f_issuance,
|
|
f_mortgage_rate,
|
|
f_overdue,
|
|
f_rate,
|
|
renormalize_contributions,
|
|
segment_steepness,
|
|
)
|
|
from app.services.forecasting.macro_series import (
|
|
MonthlyMacro,
|
|
classify_regime,
|
|
get_monthly_macro,
|
|
is_confounded_window,
|
|
macro_at_lag,
|
|
)
|
|
from app.services.forecasting.rate_sensitivity import (
|
|
RateSensitivity,
|
|
best_lag,
|
|
compute_rate_sensitivity,
|
|
ols_slope_r2,
|
|
shrink,
|
|
)
|
|
from app.services.forecasting.sales_series import (
|
|
SalesSeries,
|
|
SegmentSpec,
|
|
build_sales_series,
|
|
fill_month_grid,
|
|
log_diff,
|
|
price_bucket_of,
|
|
room_area_bucket_of,
|
|
)
|
|
|
|
__all__ = [
|
|
"DemandNormalization",
|
|
"MacroCoefficient",
|
|
"MonthlyMacro",
|
|
"RateSensitivity",
|
|
"SalesSeries",
|
|
"SegmentSpec",
|
|
"assemble_coefficient",
|
|
"best_lag",
|
|
"build_sales_series",
|
|
"classify_regime",
|
|
"compute_demand_normalization",
|
|
"compute_macro_coefficient",
|
|
"compute_rate_sensitivity",
|
|
"f_issuance",
|
|
"f_mortgage_rate",
|
|
"f_overdue",
|
|
"f_rate",
|
|
"fill_month_grid",
|
|
"get_monthly_macro",
|
|
"is_confounded_window",
|
|
"log_diff",
|
|
"macro_at_lag",
|
|
"normalization_factor",
|
|
"ols_slope_r2",
|
|
"price_bucket_of",
|
|
"renormalize_contributions",
|
|
"room_area_bucket_of",
|
|
"segment_steepness",
|
|
"shrink",
|
|
]
|