#!/usr/bin/env python3
"""採点。正解は truth.json。
(1) 項目の正答: APIが返したJSONを、項目ごとに正解と突き合わせる（全97項目×条件）
    見出し項目 … 注文番号・注文日・納期・発注側・受注側・合計金額・受け取り方法・備考の要点
    明細項目   … 品番・品名・数量（様式1は単価・金額、様式2は単位も）
    表記は NFKC でそろえ、空白を除き、×とx・全角半角・カンマの違いは誤りにしない。それ以外は1文字違えば誤り
(2) 文字として読めたか: 社名・伝票番号・品番・品名の46項目が、出力のどこかに出ているか（従来型OCRも同じ物差しで見るため）
使い方: python3 score.py            → 表を出力し results/_scores.json に保存"""
import json, os, re, glob, unicodedata
HERE = os.path.dirname(os.path.abspath(__file__))
T = json.load(open(os.path.join(HERE, 'truth.json')))
CONDS = ['scan', 'fax', 'photo']

def norm(s):
    if s is None: return ''
    s = unicodedata.normalize('NFKC', str(s)).lower()
    s = re.sub(r'[\s,，、]', '', s)
    s = s.replace('×', 'x').replace('✕', 'x').replace('‐', '-').replace('−', '-').replace('–', '-').replace('ー', 'ー')
    return s

def num(v):
    if v is None: return None
    m = re.sub(r'[^\d.]', '', unicodedata.normalize('NFKC', str(v)))
    try: return float(m) if m else None
    except ValueError: return None

def company(s):
    return re.sub(r'(御中|様|行|受注センター)$', '', re.sub(r'(御中|様|行)$', '', norm(s)))

def fields(t):
    """(項目名, 種類, 正解) の一覧"""
    out = []
    for k in ('doc_no', 'order_date', 'delivery_date', 'buyer', 'seller', 'total_amount', 'delivery_method'):
        if k in t: out.append((k, k, t[k]))
    if 'note_key' in t: out.append(('note', 'note', t['note_key']))
    for i, it in enumerate(t['items']):
        for k in ('code', 'name', 'quantity', 'unit', 'unit_price', 'amount'):
            if k in it: out.append((f'items[{i}].{k}', k, it[k]))
    return out

def get(pred, name):
    if name == 'note': return ' '.join(map(str, pred.get('notes') or [])) if isinstance(pred.get('notes'), list) else str(pred.get('notes') or '')
    m = re.match(r'items\[(\d+)\]\.(\w+)', name)
    if m:
        items = pred.get('items') or []
        i = int(m.group(1))
        return items[i].get(m.group(2)) if i < len(items) and isinstance(items[i], dict) else None
    return pred.get(name)

def ok(kind, truth, got):
    if kind in ('quantity', 'unit_price', 'amount', 'total_amount'): return num(got) == float(truth)
    if kind in ('buyer', 'seller'): return company(got) == company(truth)
    if kind in ('note', 'delivery_method'): return norm(truth) in norm(got)
    return norm(got) == norm(truth)

def load_json(p):
    s = open(p).read().strip()
    s = re.sub(r'^```(?:json)?|```$', '', s, flags=re.M).strip()
    try:
        d = json.loads(s)
        return d[0] if isinstance(d, list) and d else d
    except Exception:
        return {}

res = {}
for p in sorted(glob.glob(os.path.join(HERE, 'results', '*_form*'))):
    b = os.path.basename(p); m = re.match(r'(.+)_(form\d)_(scan|fax|photo)\.(json|txt)$', b)
    if not m: continue
    eng, form, cond, ext = m.groups(); t = T[form]; fs = fields(t)
    r = res.setdefault(eng, {}).setdefault(cond, {'field_ok': 0, 'field_n': 0, 'text_ok': 0, 'text_n': 0, 'miss': []})
    raw = norm(open(p).read().replace('\\u', ''))
    if ext == 'json':
        raw = norm(json.dumps(load_json(p), ensure_ascii=False))
        pred = load_json(p)
        for name, kind, truth in fs:
            g = get(pred, name); good = ok(kind, truth, g)
            r['field_n'] += 1; r['field_ok'] += good
            if not good: r['miss'].append(f'{form}.{name}: 正解={truth} / 出力={g}')
    for name, kind, truth in fs:
        if kind in ('doc_no', 'buyer', 'seller', 'code', 'name'):
            r['text_n'] += 1; hit = norm(truth) in raw; r['text_ok'] += hit
            if not hit and ext == 'txt': r['miss'].append(f'{form}.{name}: {truth}')
json.dump(res, open(os.path.join(HERE, 'results', '_scores.json'), 'w'), ensure_ascii=False, indent=1)
print('エンジン'.ljust(28), *[f'{c:>22}' for c in CONDS])
for eng, d in res.items():
    cells = []
    for c in CONDS:
        r = d.get(c)
        cells.append((f"項目{r['field_ok']}/{r['field_n']} " if r['field_n'] else '').rjust(12) + f"文字{r['text_ok']}/{r['text_n']}".rjust(10))
    print(eng.ljust(28), *cells)
print()
for eng, d in res.items():
    for c in CONDS:
        for m in d[c]['miss']:
            if d[c]['field_n']: print(f'[{eng} {c}] {m}')
