Winner models, code, notebooks, data and results

2b9ed19 verified 4 days ago

4.18 kB

"""
Shared configuration for the Building-Age-Classification hackathon
(Madrid -> Amsterdam transfer).
Everything that both the GBDT track and the LLM track need lives here so the
two never disagree about class boundaries, feature names or file paths.
"""
from __future__ import annotations

import os
from pathlib import Path

# --------------------------------------------------------------------------- #
# 1. Paths  (work on a laptop, on Kaggle and on Colab without edits)
# --------------------------------------------------------------------------- #

def _detect_platform() -> str:
    if os.path.exists("/kaggle/working"):
        return "kaggle"
    try:                                    # noqa: SIM105
        import google.colab  # noqa: F401
        return "colab"
    except Exception:
        return "local"


PLATFORM = _detect_platform()

if PLATFORM == "kaggle":
    PROJECT_ROOT = Path("/kaggle/working")
    DATA_DIR = Path("/kaggle/input")        # the attached dataset lives here
elif PLATFORM == "colab":
    PROJECT_ROOT = Path("/content/hack4dev")
    DATA_DIR = PROJECT_ROOT / "data"
else:
    PROJECT_ROOT = Path(__file__).resolve().parent.parent
    DATA_DIR = PROJECT_ROOT / "data"

RAW_DIR = DATA_DIR / "raw"
PROC_DIR = DATA_DIR / "processed"
REPORT_DIR = PROJECT_ROOT / "results"     # "reports" in the original project
for _d in (PROC_DIR, REPORT_DIR):
    try:
        _d.mkdir(parents=True, exist_ok=True)
    except Exception:
        pass

RANDOM_SEED = 42

# --------------------------------------------------------------------------- #
# Hugging Face settings for the LLM track (notebooks 2, 4, 5). That track is
# not part of this release; nothing here uses these names.
# --------------------------------------------------------------------------- #
HF_USER = os.environ.get("HF_USER", "Karar-git")
DATASET_REPO = f"{HF_USER}/building-age-prompts"
MODEL_REPO_PREFIX = f"{HF_USER}/"


def hf_ready() -> bool:
    return bool(HF_USER) and HF_USER != "YOUR_HF_USERNAME"

# --------------------------------------------------------------------------- #
# 2. The task definition (from the organisers' Notebook 1 / 3)
# --------------------------------------------------------------------------- #

BASE_BANDS = ["Blue", "Green", "Red", "NIR", "SWIR1", "SWIR2"]
INDEX_COLS = ["NDVI", "NDBI", "UI", "MNDWI", "BSI"]

# Landsat Collection-2 Level-2 surface-reflectance scaling: SR = DN*2.75e-5 - 0.2
SR_SCALE, SR_OFFSET = 2.75e-5, -0.2

BLUE_MAX = 15_000          # haze threshold from the organisers' Notebook 2
EARLY_CUTOFF = 2003        # early period = 1984..2003
LATE_START = 2004          # late period  = 2004..
PID = ["px_key", "py_key"]

# City-specific boundary between class 1 and class 2
CLASS_EVENTS = {"Amsterdam": 1945, "Madrid": 1960}

CLASS_LABELS = {
    "Amsterdam": {1: "before 1945", 2: "1945-1984", 3: "1984-2004", 4: "2004-2024"},
    "Madrid":    {1: "before 1960", 2: "1960-1984", 3: "1984-2004", 4: "2004-2024"},
}

# Published class shares (organisers' guide) - used only for sanity checks,
# never to fit anything.
PUBLISHED_SHARES = {"Madrid": {2: 0.356}, "Amsterdam": {4: 0.111}}

# --------------------------------------------------------------------------- #
# 3. The official 60 features, in the official order
# --------------------------------------------------------------------------- #

FEAT_OVERALL = [f"{b}_mean" for b in BASE_BANDS] + [f"{b}_std" for b in BASE_BANDS]
FEAT_INDICES = [f"{i}_mean" for i in INDEX_COLS] + [f"{i}_std" for i in INDEX_COLS]
FEAT_EARLY   = [f"{b}_early_mean" for b in BASE_BANDS] + [f"{b}_early_std" for b in BASE_BANDS]
FEAT_LATE    = [f"{b}_late_mean" for b in BASE_BANDS] + [f"{b}_late_std" for b in BASE_BANDS]
FEAT_YOY     = [f"d_{b}_mean" for b in BASE_BANDS] + [f"d_{b}_std" for b in BASE_BANDS]
FEAT_IND     = ["has_early_data", "has_late_data"]

FEATURE_COLS_60 = FEAT_OVERALL + FEAT_INDICES + FEAT_EARLY + FEAT_LATE + FEAT_YOY + FEAT_IND

FEAT_GROUPS_60 = {
    "overall": FEAT_OVERALL,
    "indices": FEAT_INDICES,
    "early": FEAT_EARLY,
    "late": FEAT_LATE,
    "yoy": FEAT_YOY,
    "indicators": FEAT_IND,
}

assert len(FEATURE_COLS_60) == 60, len(FEATURE_COLS_60)

	"""
	Shared configuration for the Building-Age-Classification hackathon
	(Madrid -> Amsterdam transfer).

	Everything that both the GBDT track and the LLM track need lives here so the
	two never disagree about class boundaries, feature names or file paths.
	"""
	from __future__ import annotations

	import os
	from pathlib import Path

	# --------------------------------------------------------------------------- #
	# 1. Paths (work on a laptop, on Kaggle and on Colab without edits)
	# --------------------------------------------------------------------------- #

	def _detect_platform() -> str:
	if os.path.exists("/kaggle/working"):
	return "kaggle"
	try: # noqa: SIM105
	import google.colab # noqa: F401
	return "colab"
	except Exception:
	return "local"


	PLATFORM = _detect_platform()

	if PLATFORM == "kaggle":
	PROJECT_ROOT = Path("/kaggle/working")
	DATA_DIR = Path("/kaggle/input") # the attached dataset lives here
	elif PLATFORM == "colab":
	PROJECT_ROOT = Path("/content/hack4dev")
	DATA_DIR = PROJECT_ROOT / "data"
	else:
	PROJECT_ROOT = Path(__file__).resolve().parent.parent
	DATA_DIR = PROJECT_ROOT / "data"

	RAW_DIR = DATA_DIR / "raw"
	PROC_DIR = DATA_DIR / "processed"
	REPORT_DIR = PROJECT_ROOT / "results" # "reports" in the original project
	for _d in (PROC_DIR, REPORT_DIR):
	try:
	_d.mkdir(parents=True, exist_ok=True)
	except Exception:
	pass

	RANDOM_SEED = 42

	# --------------------------------------------------------------------------- #
	# Hugging Face settings for the LLM track (notebooks 2, 4, 5). That track is
	# not part of this release; nothing here uses these names.
	# --------------------------------------------------------------------------- #
	HF_USER = os.environ.get("HF_USER", "Karar-git")
	DATASET_REPO = f"{HF_USER}/building-age-prompts"
	MODEL_REPO_PREFIX = f"{HF_USER}/"


	def hf_ready() -> bool:
	return bool(HF_USER) and HF_USER != "YOUR_HF_USERNAME"

	# --------------------------------------------------------------------------- #
	# 2. The task definition (from the organisers' Notebook 1 / 3)
	# --------------------------------------------------------------------------- #

	BASE_BANDS = ["Blue", "Green", "Red", "NIR", "SWIR1", "SWIR2"]
	INDEX_COLS = ["NDVI", "NDBI", "UI", "MNDWI", "BSI"]

	# Landsat Collection-2 Level-2 surface-reflectance scaling: SR = DN*2.75e-5 - 0.2
	SR_SCALE, SR_OFFSET = 2.75e-5, -0.2

	BLUE_MAX = 15_000 # haze threshold from the organisers' Notebook 2
	EARLY_CUTOFF = 2003 # early period = 1984..2003
	LATE_START = 2004 # late period = 2004..
	PID = ["px_key", "py_key"]

	# City-specific boundary between class 1 and class 2
	CLASS_EVENTS = {"Amsterdam": 1945, "Madrid": 1960}

	CLASS_LABELS = {
	"Amsterdam": {1: "before 1945", 2: "1945-1984", 3: "1984-2004", 4: "2004-2024"},
	"Madrid": {1: "before 1960", 2: "1960-1984", 3: "1984-2004", 4: "2004-2024"},
	}

	# Published class shares (organisers' guide) - used only for sanity checks,
	# never to fit anything.
	PUBLISHED_SHARES = {"Madrid": {2: 0.356}, "Amsterdam": {4: 0.111}}

	# --------------------------------------------------------------------------- #
	# 3. The official 60 features, in the official order
	# --------------------------------------------------------------------------- #

	FEAT_OVERALL = [f"{b}_mean" for b in BASE_BANDS] + [f"{b}_std" for b in BASE_BANDS]
	FEAT_INDICES = [f"{i}_mean" for i in INDEX_COLS] + [f"{i}_std" for i in INDEX_COLS]
	FEAT_EARLY = [f"{b}_early_mean" for b in BASE_BANDS] + [f"{b}_early_std" for b in BASE_BANDS]
	FEAT_LATE = [f"{b}_late_mean" for b in BASE_BANDS] + [f"{b}_late_std" for b in BASE_BANDS]
	FEAT_YOY = [f"d_{b}_mean" for b in BASE_BANDS] + [f"d_{b}_std" for b in BASE_BANDS]
	FEAT_IND = ["has_early_data", "has_late_data"]

	FEATURE_COLS_60 = FEAT_OVERALL + FEAT_INDICES + FEAT_EARLY + FEAT_LATE + FEAT_YOY + FEAT_IND

	FEAT_GROUPS_60 = {
	"overall": FEAT_OVERALL,
	"indices": FEAT_INDICES,
	"early": FEAT_EARLY,
	"late": FEAT_LATE,
	"yoy": FEAT_YOY,
	"indicators": FEAT_IND,
	}

	assert len(FEATURE_COLS_60) == 60, len(FEATURE_COLS_60)