Datasets:
Download source/server/tasks.py from theykk/ledger-check: direct link, hf CLI and curl.
- Browser
- Download file 2.78 kB
-
https://huggingface.co/datasets/theykk/ledger-check/resolve/main/source/server/tasks.py
- Command line
-
hf download hf://datasets/theykk/ledger-check/source/server/tasks.py
-
curl -L -o tasks.py https://huggingface.co/datasets/theykk/ledger-check/resolve/main/source/server/tasks.py
2.78 kB
| """Deterministic ledger tasks. Each task_id -> (ledger csv, question, expected answer).""" | |
| import random | |
| KINDS = ["balance", "category_total", "duplicate", "largest_expense", "net_flow", "unmatched"] | |
| # The server accepts any "<kind>-<n>" with n >= 1 (unbounded task pool); the | |
| # submission lists 50 of them (the arena maximum). | |
| TASK_IDS = [f"{k}-{i}" for k in KINDS for i in range(1, 9)] + ["balance-9", "category_total-9"] | |
| ACCOUNTS = ["rent", "salary", "food", "travel", "tools", "tax"] | |
| def build(task_id: str): | |
| kind, idx = task_id.rsplit("-", 1) | |
| if kind not in KINDS: | |
| raise KeyError(task_id) | |
| rng = random.Random(f"ledger-{task_id}") | |
| n = 8 + 2 * (int(idx) % 12) | |
| rows = [] | |
| for i in range(1, n + 1): | |
| acct = rng.choice(ACCOUNTS) | |
| amt = round(rng.uniform(5, 900), 2) | |
| if acct == "salary": | |
| amt = round(rng.uniform(1000, 3000), 2) | |
| else: | |
| amt = -amt | |
| rows.append([f"T{i:02d}", acct, amt]) | |
| if kind == "balance": | |
| opening = round(rng.uniform(500, 5000), 2) | |
| q = f"Opening balance is {opening:.2f}. What is the closing balance after all entries? Answer with a number, 2 decimals." | |
| ans = f"{opening + sum(r[2] for r in rows):.2f}" | |
| elif kind == "category_total": | |
| acct = rng.choice([r[1] for r in rows]) | |
| q = f"What is the total (signed sum) of all entries in account '{acct}'? Answer with a number, 2 decimals." | |
| ans = f"{sum(r[2] for r in rows if r[1] == acct):.2f}" | |
| elif kind == "duplicate": | |
| j = rng.randrange(2, n) | |
| dup = list(rows[rng.randrange(0, j)]) | |
| dup[0] = f"T{j+1:02d}x" | |
| rows.insert(j, dup) | |
| q = "Exactly one entry duplicates an earlier entry (same account and amount) with a later id. Which entry id is the duplicate?" | |
| ans = dup[0] | |
| elif kind == "largest_expense": | |
| q = "Which entry id is the largest expense (most negative amount)?" | |
| ans = min(rows, key=lambda r: r[2])[0] | |
| elif kind == "net_flow": | |
| q = "What is the sum of all positive entries minus the absolute sum of all negative entries (i.e. net flow)? Answer with a number, 2 decimals." | |
| ans = f"{sum(r[2] for r in rows):.2f}" | |
| else: # unmatched | |
| k = rng.randrange(0, n) | |
| rows[k][1] = "unknown" | |
| q = "Exactly one entry has account 'unknown'. Which entry id is it?" | |
| ans = rows[k][0] | |
| csv = "id,account,amount\n" + "\n".join(f"{r[0]},{r[1]},{r[2]:.2f}" for r in rows) | |
| return csv, q, ans | |
| def score(expected: str, given: str) -> float: | |
| g = given.strip().strip("'\"").replace(",", "") | |
| try: | |
| return 1.0 if abs(float(g) - float(expected)) < 0.005 else 0.0 | |
| except ValueError: | |
| return 1.0 if g.lower() == expected.lower() else 0.0 | |