"""The two simple baselines every tool should beat. Both use only what's known at the capture minute. Players the capture-time consensus says are ruled out get 0, so Beat-the-Baseline measures projection skill rather than just knowing who is out. """ from __future__ import annotations from pathlib import Path import pandas as pd from .names import canon_team, normalize_name HISTORY_COLS = ["date", "name_key", "team", "fpts"] def salary_rate(sal: pd.DataFrame) -> float: """The slate's points per $1K, from DraftKings' listed averages (known before lock).""" known = sal[sal["avg_ppg"] > 0] if known.empty: return 0.0 return float(known["avg_ppg"].sum() / (known["salary"].sum() / 1000.0)) def salary_baseline(sal: pd.DataFrame) -> pd.Series: rate = salary_rate(sal) return pd.Series((sal["salary"] / 1000.0 * rate).to_numpy(), index=sal["dk_id"], name="proj") def load_history(path: Path) -> pd.DataFrame: if Path(path).exists(): h = pd.read_csv(path, dtype={"name_key": str, "team": str, "date": str}) h["fpts"] = pd.to_numeric(h["fpts"], errors="coerce") return h.dropna(subset=["fpts"]) return pd.DataFrame(columns=HISTORY_COLS) def append_history(path: Path, rows: pd.DataFrame) -> None: """Add one slate's games. Regrading a slate replaces its date's rows.""" hist = load_history(path) dates = set(rows["date"].unique()) hist = hist[~hist["date"].isin(dates)] out = pd.concat([hist, rows[HISTORY_COLS]], ignore_index=True) if len(hist) else rows[HISTORY_COLS] path = Path(path) path.parent.mkdir(parents=True, exist_ok=True) out.sort_values(["date", "name_key"]).to_csv(path, index=False) def import_history(path: Path, source_csv: Path, sport: str) -> int: """Backfill games from before DFS Benchmark started (e.g. last season), so recent form can count last season when a player has fewer than the needed games. source_csv needs columns: date (YYYY-MM-DD), name, team, fpts (DraftKings points). Dates already in the history are left alone. Returns rows added. """ src = pd.read_csv(source_csv, dtype=str) src.columns = [c.strip().lower() for c in src.columns] missing = {"date", "name", "team", "fpts"} - set(src.columns) if missing: raise ValueError(f"{source_csv}: missing columns {', '.join(sorted(missing))}") rows = pd.DataFrame({ "date": src["date"].str.slice(0, 10), "name_key": src["name"].map(normalize_name), "team": src["team"].map(lambda t: canon_team(t, sport)), "fpts": pd.to_numeric(src["fpts"], errors="coerce"), }).dropna(subset=["fpts"]) hist = load_history(path) rows = rows[~rows["date"].isin(set(hist["date"]))] if rows.empty: return 0 out = pd.concat([hist, rows], ignore_index=True) if len(hist) else rows path = Path(path) path.parent.mkdir(parents=True, exist_ok=True) out.sort_values(["date", "name_key"]).to_csv(path, index=False) return int(len(rows)) def recent_form_baseline(sal: pd.DataFrame, history: pd.DataFrame, n_games: int, before: str) -> tuple[pd.Series, dict]: """Average DraftKings points over each player's last n games. Until a player has n games on record, DraftKings' listed average is used, then the salary baseline. Returns (projections, counts of which rule was used). Two players with the same name on one slate are told apart by team. """ sb = salary_baseline(sal) by_name, by_name_team = {}, {} if history is not None and len(history): h = history[history["date"] < before].sort_values("date") for keys, store in (("name_key", by_name), (["name_key", "team"], by_name_team)): g = h.groupby(keys)["fpts"] counts = g.size() means = g.apply(lambda s: s.tail(n_games).mean()) for k in means.index: if counts[k] >= n_games: store[k] = float(means[k]) dup_names = set(sal["name_key"][sal["name_key"].duplicated(keep=False)]) out, used = {}, {"history": 0, "dk_average": 0, "salary": 0} for rec in sal.itertuples(): val = by_name_team.get((rec.name_key, rec.team)) if rec.name_key in dup_names else by_name.get(rec.name_key) if val is not None: out[rec.dk_id] = val used["history"] += 1 elif rec.avg_ppg and rec.avg_ppg > 0: out[rec.dk_id] = float(rec.avg_ppg) used["dk_average"] += 1 else: out[rec.dk_id] = float(sb[rec.dk_id]) used["salary"] += 1 return pd.Series(out, name="proj"), used