""" config.py -- every notebook reads its settings from this file. Edit here, not in the notebooks. """ import os from datetime import date # ============================================================================ # CREDENTIALS # ============================================================================ # NASA token, expires after 60 days. Get one at # https://ladsweb.modaps.eosdis.nasa.gov -> My Account -> Generate Token NASA_TOKEN = "" # FRED API key, free at https://fredaccount.stlouisfed.org/apikeys FRED_API_KEY = "" # ============================================================================ # COUNTRY # ============================================================================ COUNTRY = "Bolivia" # must be a key of COUNTRIES COUNTRIES = { "Bolivia": dict(iso2="BO", iso3="BOL", bbox=(-69.65, -22.91, -57.45, -9.67), tiles=["h11v09", "h11v10", "h12v10", "h11v11", "h12v11"]), } if COUNTRY not in COUNTRIES: raise KeyError(f"{COUNTRY!r} is not in COUNTRIES. Known: {sorted(COUNTRIES)}") _C = COUNTRIES[COUNTRY] CODE = _C["iso2"] # "BO" CODE_LC = CODE.lower() # "bo" GADM_ISO3 = _C["iso3"] # "BOL" BBOX = _C["bbox"] # (lon_min, lat_min, lon_max, lat_max) # ---- Night-time light tiles (VIIRS Black Marble) --------------------------- TILES = list(_C["tiles"]) TILES_OVERRIDE = ["h11v09", "h11v10", "h12v10", "h11v11"] # None = use TILES if TILES_OVERRIDE: # (h12v11 is 0.02% of Bolivia) TILES = list(TILES_OVERRIDE) DOWNLOAD_TILES = None # subset of TILES to download; None = all # ---- File names ------------------------------------------------------------ GDP_TICKER = "RGDP0000" CPI_TICKER = "MCPI0000" CPI_BASE_YEAR = 2016 COORDINATES_XLSX = "Coordinates.xlsx" # NTL sites (lat/lon), in raw/ NTL_OUTPUT_CSV = "ntl_data.csv" # daily NTL per site, in raw/ NTL_SHAPE_CSV = "ntl_shape.csv" # daily NTL, country mean, in raw/ DATA_CSV = "data.csv" METADATA_CSV = "meta.csv" # ============================================================================ # NIGHT-TIME LIGHTS (act3_ntl) # ============================================================================ # Window to download and analyse. About 4 files (~100 MB) per day. # "2026-01-01 2026-01-09" -> from - to # "2026-01-01" -> from that date to the latest available # None -> from NTL_FIRST_DATE to the latest available NTL_DATE_RANGE = "2026-01-01 2026-01-09" # Safety limits, applied even when NTL_DATE_RANGE is None. NTL_FIRST_DATE = "2016-11-01" # never download anything earlier NTL_MAX_FILES = 4000 # refuse a run above this (~100 GB); None = no limit # Event marked on the series and centred in the before/after maps. NTL_EVENT_DATE = "2026-01-04" # e.g. "2026-01-20"; None = first/last day of the window NTL_EVENT_LABEL = "Ejemplo" # e.g. "paro de combustible" NTL_EVENT_WINDOW = 2 # days before/after the event shown on the maps NTL_DRAW_MAPS = True # False = skip the maps (they re-open the .h5 tiles) def ntl_window(): """(start, end) of the NTL window as datetimes; end None = latest available.""" from datetime import datetime floor = datetime.strptime(NTL_FIRST_DATE, "%Y-%m-%d") if NTL_FIRST_DATE else None if not NTL_DATE_RANGE: return floor, None parts = NTL_DATE_RANGE.split() if len(parts) == 1: start, end = datetime.strptime(parts[0], "%Y-%m-%d"), None elif len(parts) == 2: start = datetime.strptime(parts[0], "%Y-%m-%d") end = datetime.strptime(parts[1], "%Y-%m-%d") else: raise ValueError("NTL_DATE_RANGE must be None, one date, or two dates " "separated by a space; got %r" % (NTL_DATE_RANGE,)) if floor and start < floor: start = floor return start, end # ============================================================================ # NOWCAST # ============================================================================ DESIRED_DATE = "2026-06-01" # quarter to nowcast (last month of the quarter) DATE_STR = date.today().strftime("%Y%m%d") # results go to output// TRAIN_START = "2015-03-01" # training sample start TEST_START = "2023-06-01" # test sample start LAGS = list(range(-2, 3)) # vintages -2..+2 (backcast .. forecast) LARS_FACTORS = 1 # PCA components in the LARS benchmark SEASONAL_FILTER = 0 # 0 = classical (fast), 1 = X-13ARIMA-SEATS (slow) X13_PATH = None # folder with x13as; None = look on PATH RUN = { # models to fit: 1 = on, 0 = off "ols": 1, "olsr": 1, "enet": 1, "lasso": 1, "gbt": 1, "dt": 1, "rf": 1, "lstm": 1, } ITER = 10 # ensemble size for DT / RF / GBT (production: 100) ITER_LSTM = 10 # LSTM ensemble size EPOCH_LSTM = 50 # LSTM training epochs # ---- Advanced options: all off for a plain nowcast ------------------------- SCALE_LINEAR = 1 # scale predictors for Ridge / ENET / LASSO (keep on) SHOCK_FEATURES = 0 ROLLING_LARS = 0 SELECTION_CV = 0 REGIME_ENSEMBLE = 0 WATCHLIST = 0 DISPERSION_BANDS = 0 SAMPLE_WEIGHT_DECAY = None NWCST_SEED = 42 NWCST_FILL = "mean" # ragged-edge fill: "mean" or "ffill" RIDGE_ALPHAS = [0.1, 1, 10, 50, 100] MAX_WORKERS = None # None = based on the number of CPUs SHOCK_WINSOR_Z = 3.0 SHOCK_BREAK_Z = -3.0 SHOCK_MIN_BROKEN = 4 ROLLING_MAX_VARS = 99 CV_MIN_TRAIN = 16 CV_STEP = 2 CV_EVAL_SIZE = 4 LARS_SIZE_RANGE = (10, 50) SHOCK_DATES = [] LOO_SHOCK_DATES = [] SHOCK_TARGET_DATE = None ORACLE_KS = [] ORACLE_ESTIMATORS = ["ENET", "LASSO"] ORACLE_VINTAGE = 1 ORACLE_KS_TREE = [] ORACLE_TREE_ESTIMATORS = ["DT", "RF", "GBT"] ORACLE_TREE_VINTAGE = 2 # ============================================================================ # PATHS AND HELPERS -- normally no need to edit below # ============================================================================ GADM_ZIP = "gadm41_" + GADM_ISO3 + "_shp.zip" PROJECT_ROOT = ".." # project root, relative to this file NL_ROOT = None # folder with the .h5 tiles; None = /hfiles def tiles_from_bbox(bbox=None): """VIIRS tiles touched by a bounding box (first guess for a new country).""" import math lo, la, hi, ha = bbox or BBOX h0, h1 = math.floor((lo + 180) / 10), math.floor((hi + 180) / 10) v0, v1 = math.floor((90 - ha) / 10), math.floor((90 - la) / 10) return [f"h{h:02d}v{v:02d}" for v in range(v0, v1 + 1) for h in range(h0, h1 + 1)] def refresh_registry(shape_dir=None, countries=None): """Recompute bbox + tiles from the GADM border; prints lines for COUNTRIES. conda run -n geo python -c "import config; config.refresh_registry()" """ import urllib.request import geopandas as gpd from shapely.geometry import box shape_dir = shape_dir or os.path.join(os.getcwd(), "shapes") os.makedirs(shape_dir, exist_ok=True) out = {} for name in (countries or COUNTRIES): iso3 = COUNTRIES[name]["iso3"] zp = os.path.join(shape_dir, f"gadm41_{iso3}_shp.zip") if not os.path.exists(zp): urllib.request.urlretrieve( f"https://geodata.ucdavis.edu/gadm/gadm4.1/shp/gadm41_{iso3}_shp.zip", zp) g = gpd.read_file(f"zip://{zp}!gadm41_{iso3}_0.shp") geom = g.geometry.union_all() if hasattr(g.geometry, "union_all") else g.geometry.unary_union b = geom.bounds cand = tiles_from_bbox(b) exact = [t for t in cand if geom.intersects(box(10 * int(t[1:3]) - 180, 90 - 10 * int(t[4:6]) - 10, 10 * int(t[1:3]) - 170, 90 - 10 * int(t[4:6])))] out[name] = (b, exact) drop = set(cand) - set(exact) print(f' {name!r:24}: dict(iso2="{COUNTRIES[name]["iso2"]}", iso3="{iso3}", ' f'bbox=({b[0]:.2f}, {b[1]:.2f}, {b[2]:.2f}, {b[3]:.2f}),') print(f'{"":30}tiles={exact}),' + (f' # bbox rule also suggested {sorted(drop)} -- not touched' if drop else "")) return out def download_tiles(): """Tiles to download: DOWNLOAD_TILES if set, else TILES.""" return list(DOWNLOAD_TILES) if DOWNLOAD_TILES else list(TILES) def derive_paths(d, make=True): """Project folders, given d = the folder holding this file. Creates missing ones.""" d = os.path.abspath(d) root = os.path.abspath(os.path.join(d, PROJECT_ROOT)) up = lambda *p: os.path.join(root, *p) P = { "code": d, "raw": up("raw"), "data": up("data"), "output": up("output"), "ntl_tmp": up("ntl_tmp"), "nl_root": os.path.abspath(os.path.join(d, NL_ROOT)) if NL_ROOT else up("hfiles"), } if make: for k, p in P.items(): if k != "code": os.makedirs(p, exist_ok=True) return P