import warnings; warnings.filterwarnings("ignore")
"""Shared paths and helpers for the 05_Predictions layer (Binzagr PoV)."""
import os, numpy as np, pandas as pd
ROOT = os.environ.get('BZ_ROOT', '/mnt/user-data/outputs/Binzagr_Data')
CLEAN = os.path.join(ROOT, '02_Cleaned', 'data')
SIM = os.path.join(ROOT, '03_Simulated', 'data')
OUT = os.path.join(ROOT, '05_Predictions')
WORK = os.environ.get('BZ_WORK', '/home/claude/binzagr/pred/work')
os.makedirs(WORK, exist_ok=True)

DATA_END = pd.Timestamp('2026-10-20')          # last sales day (Tuesday)
WEEK_FREQ = 'W-TUE'                             # weeks run Wednesday -> Tuesday, so the last week ends on DATA_END
FIRST_FULL_WEEK_END = pd.Timestamp('2026-03-10')
TRAIN_END = pd.Timestamp('2026-09-22')          # forecast models trained to here (wireframe: "trained to 22 Sep 2026")
HOLDOUT_WEEKS = [pd.Timestamp(d) for d in ('2026-09-29', '2026-10-06', '2026-10-13', '2026-10-20')]
RECO_CUTOFF = pd.Timestamp('2026-08-25')        # next-best SKU picks made 25 Aug 2026, tested on the 8 unseen weeks after it
NEXT_WEEK = pd.date_range('2026-10-21', '2026-10-27')   # "next week" for van plans


def out_dir(sub):
    d = os.path.join(OUT, sub)
    os.makedirs(d, exist_ok=True)
    return d


def write_csv(df, sub, name, float_fmt=None):
    """UTF-8 with BOM (Arabic names display in Excel), ISO dates, no index."""
    p = os.path.join(out_dir(sub), name)
    df = df.copy()
    for c in df.columns:
        if pd.api.types.is_datetime64_any_dtype(df[c]):
            df[c] = df[c].dt.strftime('%Y-%m-%d')
    df.to_csv(p, index=False, encoding='utf-8-sig', float_format=float_fmt)
    print(f'  wrote {sub}/{name}: {len(df):,} rows x {df.shape[1]} cols')
    return p


def week_end(d):
    """Tuesday that closes the Wed-Tue week containing date d."""
    d = pd.to_datetime(d)
    return (d + pd.to_timedelta((1 - d.dt.weekday) % 7, unit='D')).dt.normalize()


def wape(y, f):
    y = np.asarray(y, float); f = np.asarray(f, float)
    s = np.abs(y).sum()
    return float(np.abs(y - f).sum() / s) if s > 0 else np.nan


def outlet_id(dist, cust):
    return dist.astype(str) + '-' + cust.astype(str)