be3e51bae9
Size the predicted dwelling from the geo-proximity-weighted median of the cohort's floor areas rather than the plain median: homes built together share a footprint, so a nearer neighbour's area should count for more (the same street signal #1227 already wired into age / wall / glazing). Reuses `_geo_weights` and adds `_weighted_median`, which reduces exactly to `statistics.median` under uniform weights (geo off / no target coordinates) — including the even-count midpoint average — so the MAD-minimising guarantee is preserved. Measured over the 514-target SAP-10.2 corpus (leave-one-out): floor_area MAE 10.48 -> 9.73 m² MAPE 13.2% -> 12.2% Re-baselines the n=36 fixture floor_area ceiling 11.8983 -> 12.0378 (a method change, not a loosening; the small fixture subset moved +0.14 the other way as sample noise while the population improved decisively). The ceiling still pins the new deterministic value exactly, so the tighten-only ratchet resumes. Investigation ruling out the adjacent floor-area levers (kept in the follow-up): lowering minimum_cohort (9.78-10.03, worse), hard same-form filter (10.19), mean instead of median (10.68), constant bias correction (10.47), extension-conditioning (oracle 9.50, not worth the misclassification cost) and room-in-roof conditioning/additive (RiR is a confound for large multi-part outliers — RiR area is only ~21% of total, and the increment breaks the homes already predicted exactly). Remaining cohort lever is built-form soft-weighting, gated on a denser corpus. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|---|---|---|
| .devcontainer | ||
| .github/workflows | ||
| .idea | ||
| .vscode | ||
| applications | ||
| asset_list | ||
| backend | ||
| backlog | ||
| datatypes | ||
| deployment/terraform | ||
| docs | ||
| domain | ||
| epr_data_exports | ||
| etl | ||
| harness | ||
| infrastructure | ||
| model_data/requirements | ||
| orchestration | ||
| recommendations | ||
| repositories | ||
| sap worksheets | ||
| scripts | ||
| sfr/principal_pitch | ||
| survey_report | ||
| tests | ||
| utilities | ||
| utils | ||
| .coveragerc | ||
| .dockerignore | ||
| .gitignore | ||
| __init__.py | ||
| ara_backend_design.md | ||
| BaseUtility.py | ||
| CLAUDE.md | ||
| conftest.py | ||
| CONTEXT.md | ||
| devcontainer.sh | ||
| Dockerfile.test | ||
| Dockerfile.test.dockerignore | ||
| Makefile | ||
| MEMORY.md | ||
| next_claude_prompt.txt | ||
| package-lock.json | ||
| package.json | ||
| pyproject.toml | ||
| pyrightconfig.json | ||
| pytest.ini | ||
| README.md | ||
| run_lambda_local.sh | ||
| serverless.yml | ||
| test.requirements.txt | ||
| tox.ini | ||
| UBIQUITOUS_LANGUAGE.md | ||
Model Repository
This repository contains the code pertaining to the development of the data science and machine learning products being utilised by Hestia.
The different folders in this repository relate to services that can be used independently, or can be imported and used as part of a larger application
Getting Started
Prerequisites
Dev Container Setup
This repo uses a Docker Compose-based dev container. The model-backend service joins a shared-dev Docker network so it can communicate with other local services (e.g. a frontend container) running on your machine.
VS Code users: The initializeCommand in devcontainer.json creates the shared-dev network automatically before the container starts. No manual step required — just open the repo and select Reopen in Container.
Non-VS Code / CI workflows: Run the following once before starting the container:
make dev-setup
This is idempotent and safe to re-run if the network already exists.
Folders
backend/
This folder contains the code for the fastapi backend service, which provides an interface to much of the functionality in this repository, for the frontend
model_data/
This folder contains related to the reading and preparation of assessment model data, including pulling out epc attributes
Testing
All tests can be run, against the configuration in pytest.ini running
pytest
This will run the complete panel of tests and report on coverage in the locations specified by the pytest.ini file.
To run tests in a specific service, e.g. inside of model_data, simply run
pytest --cov-config=model_data/.coveragerc --cov=model_data
This will produce the test results and coverage reports