junte/Model
1
0
Fork 0
mirror of https://github.com/Hestia-Homes/Model.git synced 2026-07-27 23:35:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Re-baseline epc-prediction gates for the #1669 ground-truth change 🟩

Browse source 
The ADR-0030 component-accuracy and expired-pairs gates re-derive the *actual*
EpcPropertyData through the same mapper the leave-one-out scorer predicts against.
#1669 (measurement_type=2 external->internal conversion, RdSAP §3.4 + Table 2)
gives the fixture's external certs sharper internal dimensions, which tips the
geo-proximity-weighted donor mode on three components by single-target amounts:
total_window_area 3.7184 -> 3.7484, door_count 0.3333 -> 0.3737, has_pv floor
0.8929 -> 0.8571 (one pair 25/28 -> 24/28).

This is a ground-truth-method change, not a prediction-logic loosening: the fix
is spec-correct, corpus MAE improved (0.571 -> 0.570) and the real-cert accuracy
pin held. Per the gate convention the affected floors/ceilings are re-baselined
to the measured values with the attribution recorded; tighten-only resumes here.
Bisected: #1666/#1667 leave the gates green; #1669 introduces all three shifts.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Khalim Conn-Kowlessar 2026-07-22 15:34:12 +00:00
parent e2404ea303
commit 6b4f883bc8
2 changed files with 21 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
tests/domain/epc_prediction/test_component_accuracy_gate.py
View file

@ -124,11 +124,20 @@ _RATE_FLOORS: dict[str, float] = {
# dimensional predictions); floor_area loosened 12.0378 -> 12.0586 as the one
# physical residual that fell (1-2 targets picking a new-build donor). See the
# _RATE_FLOORS note above.
# total_window_area 3.7184 -> 3.7484 and door_count 0.3333 -> 0.3737 re-baselined
# when the external-measurement conversion landed (#1669: measurement_type=2 certs
# now have their floor area + exposed perimeter converted external->internal per
# RdSAP 10 §3.4 + Table 2). The leave-one-out scorer re-derives the *actual*
# EpcPropertyData through the same mapper, so the sharper (internal) dimensions of
# the fixture's external certs shift the geo-proximity-weighted donor mode — a
# ground-truth-method change, not a prediction-logic loosening (spec-correct,
# corpus MAE 0.571 -> 0.570 and the real-cert pin held). The move is two
# single-target donor tips on the n=36 fixture. Tighten-only resumes from here.
_RESIDUAL_CEILINGS: dict[str, float] = {
"floor_area": 12.0586,
"total_window_area": 3.7184,
"total_window_area": 3.7484,
"building_parts": 0.1212,
"door_count": 0.3333,
"door_count": 0.3737,
}
_TOLERANCE = 1e-3

11
tests/domain/epc_prediction/test_expired_pairs_gate.py
View file

@ -55,6 +55,15 @@ _FIXTURE = (
# 28-scoreable fixture) and residual ceilings — the measured values; tighten,
# never loosen. The general (unconditioned) prediction floors live in
# test_component_accuracy_gate.py; this gate guards the CONDITIONING path.
#
# has_pv re-baselined 0.8929 -> 0.8571 when the external-measurement conversion
# landed (#1669: measurement_type=2 certs converted external->internal per RdSAP
# 10 §3.4 + Table 2). This gate re-derives the *actual* EpcPropertyData through
# the same mapper, so the sharper internal dimensions of the fixture's external
# certs shift the conditioned donor mode and one pair's has_pv agreement tips
# 25/28 -> 24/28 — a ground-truth-method change, not a prediction-logic loosening
# (spec-correct; corpus MAE improved and the real-cert pin held). Tighten-only
# resumes from here.
_CLASSIFICATION_FLOORS: dict[str, float] = {
"construction_age_band": 0.5357,
"construction_age_band_pm1": 0.8214,
@ -62,7 +71,7 @@ _CLASSIFICATION_FLOORS: dict[str, float] = {
"floor_construction": 0.8636,
"floor_insulation": 1.0000,
"has_hot_water_cylinder": 0.8214,
"has_pv": 0.8929,
"has_pv": 0.8571,
"has_room_in_roof": 0.8571,
"heating_main_category": 1.0000,
"heating_main_control": 0.6071,