"""Value-class buffers for the reorder study: runs the forecast policy with a different z per ABC value class.
Classes come from demand value in the warm-up period only (A: top 80% of value, B: next 15%, C: last 5%). Reuses stock.py's simulation."""
import json, numpy as np
src = open('/workspace/cs6/code/stock.py').read().split("rule = {b:")[0]; exec(src)
val = D[:t0].sum(0) * price; order = np.argsort(-val); cls = np.empty(K, int); cum = np.cumsum(val[order]) / val.sum(); cls[order] = np.where(cum <= .8, 0, np.where(cum <= .95, 1, 2))
out = json.load(open('/workspace/cs6/out/stock.json')); res = {}
for nm, zz in {'a1.0_b1.65_c2.33': (1.0, 1.65, 2.33), 'a1.28_b1.65_c2.33': (1.28, 1.65, 2.33), 'a0.84_b1.28_c2.05': (0.84, 1.28, 2.05), 'a1.0_b2.0_c2.5': (1.0, 2.0, 2.5), 'a1.28_b2.0_c2.5': (1.28, 2.0, 2.5)}.items():
    res[nm] = sim('model', np.array(zz)[cls])
out['classes'] = dict(sizes=np.bincount(cls).tolist(), value_share=[float(val[cls == c].sum() / val.sum()) for c in range(3)], runs=res)
json.dump(out, open('/workspace/cs6/out/stock.json', 'w'), indent=1)
