You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
405 lines
13 KiB
Python
405 lines
13 KiB
Python
"""Write all output artifacts for a backtest run."""
|
|
from __future__ import annotations
|
|
|
|
import csv
|
|
import datetime as dt
|
|
import json
|
|
from collections import defaultdict
|
|
from pathlib import Path
|
|
from typing import Any
|
|
|
|
import pyarrow as pa
|
|
import pyarrow.parquet as pq
|
|
|
|
from libs.backtest.domain import (
|
|
BacktestConfig,
|
|
DailyPortfolioState,
|
|
ExperimentManifest,
|
|
ExperimentResult,
|
|
FilledTrade,
|
|
MetricsBundle,
|
|
OpenPosition,
|
|
)
|
|
from libs.common.logging import get_logger
|
|
|
|
logger = get_logger(__name__)
|
|
|
|
|
|
def create_run_directory(output_root: str | Path, run_id: str) -> Path:
|
|
"""Create the full run directory tree. Returns the run root path."""
|
|
run_dir = Path(output_root) / run_id
|
|
for subdir in ["logs", "metrics", "artifacts", "plots", "notes"]:
|
|
(run_dir / subdir).mkdir(parents=True, exist_ok=True)
|
|
logger.info("run_directory_created", path=str(run_dir))
|
|
return run_dir
|
|
|
|
|
|
def write_metadata(
|
|
run_dir: Path,
|
|
run_id: str,
|
|
started_at: dt.datetime,
|
|
finished_at: dt.datetime,
|
|
git_hash: str,
|
|
total_trading_days: int,
|
|
total_candidates_seen: int,
|
|
total_orders_rejected: int,
|
|
split_name: str | None = None,
|
|
) -> Path:
|
|
"""Write metadata.json."""
|
|
meta = {
|
|
"run_id": run_id,
|
|
"started_at": started_at.isoformat(),
|
|
"finished_at": finished_at.isoformat(),
|
|
"elapsed_seconds": (finished_at - started_at).total_seconds(),
|
|
"git_commit_hash": git_hash,
|
|
"total_trading_days": total_trading_days,
|
|
"total_candidates_seen": total_candidates_seen,
|
|
"total_orders_rejected": total_orders_rejected,
|
|
}
|
|
if split_name is not None:
|
|
meta["split_name"] = split_name
|
|
out = run_dir / "metadata.json"
|
|
out.write_text(json.dumps(meta, indent=2))
|
|
return out
|
|
|
|
|
|
def write_metrics_summary(run_dir: Path, metrics: MetricsBundle) -> Path:
|
|
"""Write metrics/metrics_summary.json."""
|
|
out = run_dir / "metrics" / "metrics_summary.json"
|
|
out.write_text(metrics.model_dump_json(indent=2))
|
|
return out
|
|
|
|
|
|
def write_trade_blotter(
|
|
run_dir: Path,
|
|
trades: list[FilledTrade],
|
|
) -> Path | None:
|
|
"""Write artifacts/trade_blotter.parquet."""
|
|
if not trades:
|
|
return None
|
|
rows = [
|
|
{
|
|
"trade_id": t.trade_id,
|
|
"position_id": t.position_id,
|
|
"event_id": t.event_id,
|
|
"symbol": t.symbol,
|
|
"entry_date": t.entry_date.isoformat(),
|
|
"exit_date": t.exit_date.isoformat(),
|
|
"entry_price": t.entry_price,
|
|
"exit_price": t.exit_price,
|
|
"exit_reason": t.exit_reason.value,
|
|
"shares": t.shares,
|
|
"commission": t.commission,
|
|
"slippage_bps": t.slippage_bps,
|
|
"gross_pnl": t.gross_pnl,
|
|
"net_pnl": t.net_pnl,
|
|
"pnl_pct": t.pnl_pct,
|
|
"r_multiple": t.r_multiple,
|
|
"holding_days": t.holding_days,
|
|
}
|
|
for t in trades
|
|
]
|
|
out = run_dir / "artifacts" / "trade_blotter.parquet"
|
|
_write_parquet(rows, out)
|
|
return out
|
|
|
|
|
|
def write_daily_equity_curve(
|
|
run_dir: Path,
|
|
equity_curve: list[DailyPortfolioState],
|
|
) -> Path | None:
|
|
"""Write artifacts/daily_equity_curve.parquet."""
|
|
if not equity_curve:
|
|
return None
|
|
rows = [
|
|
{
|
|
"date": s.date.isoformat(),
|
|
"equity": s.equity,
|
|
"cash_available": s.cash_available,
|
|
"gross_exposure": s.gross_exposure,
|
|
"net_exposure": s.net_exposure,
|
|
"unrealized_pnl": s.unrealized_pnl,
|
|
"realized_pnl": s.realized_pnl,
|
|
"open_position_count": len(s.open_positions),
|
|
"daily_new_risk_used": s.daily_new_risk_used,
|
|
"peak_equity": s.peak_equity,
|
|
"current_drawdown_pct": s.current_drawdown_pct,
|
|
}
|
|
for s in equity_curve
|
|
]
|
|
out = run_dir / "artifacts" / "daily_equity_curve.parquet"
|
|
_write_parquet(rows, out)
|
|
return out
|
|
|
|
|
|
def write_position_timeline(
|
|
run_dir: Path,
|
|
trades: list[FilledTrade],
|
|
open_positions: list[OpenPosition] | None = None,
|
|
) -> Path | None:
|
|
"""Write artifacts/position_timeline.parquet (one row per position)."""
|
|
rows = []
|
|
for t in trades:
|
|
rows.append(
|
|
{
|
|
"position_id": t.position_id,
|
|
"event_id": t.event_id,
|
|
"symbol": t.symbol,
|
|
"entry_date": t.entry_date.isoformat(),
|
|
"exit_date": t.exit_date.isoformat(),
|
|
"entry_price": t.entry_price,
|
|
"exit_price": t.exit_price,
|
|
"exit_reason": t.exit_reason.value,
|
|
"shares": t.shares,
|
|
"net_pnl": t.net_pnl,
|
|
"r_multiple": t.r_multiple,
|
|
"holding_days": t.holding_days,
|
|
"status": "closed",
|
|
}
|
|
)
|
|
if open_positions:
|
|
for p in open_positions:
|
|
rows.append(
|
|
{
|
|
"position_id": p.position_id,
|
|
"event_id": p.plan.candidate.event_id,
|
|
"symbol": p.plan.candidate.symbol,
|
|
"entry_date": p.entry_date.isoformat(),
|
|
"exit_date": None,
|
|
"entry_price": p.entry_price,
|
|
"exit_price": None,
|
|
"exit_reason": None,
|
|
"shares": p.shares_total,
|
|
"net_pnl": None,
|
|
"r_multiple": None,
|
|
"holding_days": p.days_held,
|
|
"status": p.status.value,
|
|
}
|
|
)
|
|
if not rows:
|
|
return None
|
|
out = run_dir / "artifacts" / "position_timeline.parquet"
|
|
_write_parquet(rows, out)
|
|
return out
|
|
|
|
|
|
def write_attribution_by_event_type(
|
|
run_dir: Path,
|
|
trades: list[FilledTrade],
|
|
candidate_map: dict[str, Any],
|
|
) -> Path:
|
|
"""Write metrics/attribution_by_event_type.csv."""
|
|
bucket_data: dict[str, dict[str, float | int]] = defaultdict(
|
|
lambda: {"count": 0, "wins": 0, "net_pnl": 0.0, "avg_r": 0.0, "_r_sum": 0.0}
|
|
)
|
|
for t in trades:
|
|
cand = candidate_map.get(t.trade_id)
|
|
et = getattr(cand, "event_type", "unknown") if cand else "unknown"
|
|
d = bucket_data[et]
|
|
d["count"] = int(d["count"]) + 1
|
|
if t.net_pnl > 0:
|
|
d["wins"] = int(d["wins"]) + 1
|
|
d["net_pnl"] = float(d["net_pnl"]) + t.net_pnl
|
|
d["_r_sum"] = float(d["_r_sum"]) + t.r_multiple
|
|
|
|
out = run_dir / "metrics" / "attribution_by_event_type.csv"
|
|
with open(out, "w", newline="") as f:
|
|
writer = csv.DictWriter(
|
|
f, fieldnames=["event_type", "count", "wins", "win_rate", "net_pnl", "avg_r"]
|
|
)
|
|
writer.writeheader()
|
|
for et, d in sorted(bucket_data.items()):
|
|
count = int(d["count"])
|
|
wins = int(d["wins"])
|
|
writer.writerow(
|
|
{
|
|
"event_type": et,
|
|
"count": count,
|
|
"wins": wins,
|
|
"win_rate": wins / count if count > 0 else 0.0,
|
|
"net_pnl": round(float(d["net_pnl"]), 4),
|
|
"avg_r": round(float(d["_r_sum"]) / count if count > 0 else 0.0, 4),
|
|
}
|
|
)
|
|
return out
|
|
|
|
|
|
def write_attribution_by_sector(
|
|
run_dir: Path,
|
|
trades: list[FilledTrade],
|
|
candidate_map: dict[str, Any],
|
|
) -> Path:
|
|
"""Write metrics/attribution_by_sector.csv."""
|
|
bucket_data: dict[str, dict[str, float | int]] = defaultdict(
|
|
lambda: {"count": 0, "wins": 0, "net_pnl": 0.0, "_r_sum": 0.0}
|
|
)
|
|
for t in trades:
|
|
cand = candidate_map.get(t.trade_id)
|
|
sector = getattr(cand, "sector", "UNKNOWN") if cand else "UNKNOWN"
|
|
d = bucket_data[sector]
|
|
d["count"] = int(d["count"]) + 1
|
|
if t.net_pnl > 0:
|
|
d["wins"] = int(d["wins"]) + 1
|
|
d["net_pnl"] = float(d["net_pnl"]) + t.net_pnl
|
|
d["_r_sum"] = float(d["_r_sum"]) + t.r_multiple
|
|
|
|
out = run_dir / "metrics" / "attribution_by_sector.csv"
|
|
with open(out, "w", newline="") as f:
|
|
writer = csv.DictWriter(
|
|
f, fieldnames=["sector", "count", "wins", "win_rate", "net_pnl", "avg_r"]
|
|
)
|
|
writer.writeheader()
|
|
for sector, d in sorted(bucket_data.items()):
|
|
count = int(d["count"])
|
|
wins = int(d["wins"])
|
|
writer.writerow(
|
|
{
|
|
"sector": sector,
|
|
"count": count,
|
|
"wins": wins,
|
|
"win_rate": wins / count if count > 0 else 0.0,
|
|
"net_pnl": round(float(d["net_pnl"]), 4),
|
|
"avg_r": round(float(d["_r_sum"]) / count if count > 0 else 0.0, 4),
|
|
}
|
|
)
|
|
return out
|
|
|
|
|
|
def write_score_bucket_report(
|
|
run_dir: Path,
|
|
score_bucket_hit_rate: dict[str, float],
|
|
trades: list[FilledTrade],
|
|
candidate_map: dict[str, Any],
|
|
) -> Path:
|
|
"""Write metrics/score_bucket_report.csv."""
|
|
bucket_counts: dict[str, int] = defaultdict(int)
|
|
for t in trades:
|
|
cand = candidate_map.get(t.trade_id)
|
|
bucket = getattr(cand, "score_bucket", "unknown") if cand else "unknown"
|
|
bucket_counts[bucket] += 1
|
|
|
|
out = run_dir / "metrics" / "score_bucket_report.csv"
|
|
with open(out, "w", newline="") as f:
|
|
writer = csv.DictWriter(f, fieldnames=["score_bucket", "trade_count", "win_rate"])
|
|
writer.writeheader()
|
|
for bucket in sorted(set(list(score_bucket_hit_rate.keys()) + list(bucket_counts.keys()))):
|
|
writer.writerow(
|
|
{
|
|
"score_bucket": bucket,
|
|
"trade_count": bucket_counts.get(bucket, 0),
|
|
"win_rate": round(score_bucket_hit_rate.get(bucket, 0.0), 4),
|
|
}
|
|
)
|
|
return out
|
|
|
|
|
|
def write_plots(run_dir: Path, generate: bool = False) -> Path:
|
|
"""Create plots directory. generate=True logs a warning (matplotlib not available)."""
|
|
plots_dir = run_dir / "plots"
|
|
plots_dir.mkdir(exist_ok=True)
|
|
if generate:
|
|
logger.warning(
|
|
"plots_not_implemented",
|
|
message="generate_plots=True is a no-op; matplotlib is not in dependencies.",
|
|
)
|
|
return plots_dir
|
|
|
|
|
|
def write_run_notes(run_dir: Path, notes: str = "") -> Path:
|
|
"""Write notes/run_notes.md."""
|
|
out = run_dir / "notes" / "run_notes.md"
|
|
out.write_text(notes or "# Run Notes\n\n_No notes provided._\n")
|
|
return out
|
|
|
|
|
|
def write_all_artifacts(
|
|
run_dir: Path,
|
|
run_id: str,
|
|
manifest: ExperimentManifest,
|
|
config: BacktestConfig,
|
|
metrics: MetricsBundle,
|
|
trades: list[FilledTrade],
|
|
equity_curve: list[DailyPortfolioState],
|
|
open_positions: list[OpenPosition],
|
|
candidate_map: dict[str, Any],
|
|
started_at: dt.datetime,
|
|
finished_at: dt.datetime,
|
|
git_hash: str,
|
|
total_trading_days: int,
|
|
total_candidates_seen: int,
|
|
total_orders_rejected: int,
|
|
split_name: str | None = None,
|
|
) -> dict[str, str]:
|
|
"""Write all output files. Returns mapping of artifact_name → file_path."""
|
|
from libs.backtest.manifests import save_manifest, save_resolved_config
|
|
|
|
paths: dict[str, str] = {}
|
|
|
|
# Core files
|
|
paths["manifest"] = str(save_manifest(manifest, run_dir))
|
|
paths["resolved_config"] = str(save_resolved_config(config, run_dir))
|
|
paths["metadata"] = str(
|
|
write_metadata(
|
|
run_dir, run_id, started_at, finished_at, git_hash,
|
|
total_trading_days, total_candidates_seen, total_orders_rejected,
|
|
split_name=split_name,
|
|
)
|
|
)
|
|
|
|
# Metrics
|
|
if config.reporting.write_metrics_summary:
|
|
paths["metrics_summary"] = str(write_metrics_summary(run_dir, metrics))
|
|
paths["attribution_by_event_type"] = str(
|
|
write_attribution_by_event_type(run_dir, trades, candidate_map)
|
|
)
|
|
paths["attribution_by_sector"] = str(
|
|
write_attribution_by_sector(run_dir, trades, candidate_map)
|
|
)
|
|
paths["score_bucket_report"] = str(
|
|
write_score_bucket_report(
|
|
run_dir, metrics.score_bucket_hit_rate, trades, candidate_map
|
|
)
|
|
)
|
|
|
|
# Trade data
|
|
if config.reporting.write_trade_blotter:
|
|
p = write_trade_blotter(run_dir, trades)
|
|
if p:
|
|
paths["trade_blotter"] = str(p)
|
|
|
|
if config.reporting.write_equity_curve:
|
|
p = write_daily_equity_curve(run_dir, equity_curve)
|
|
if p:
|
|
paths["daily_equity_curve"] = str(p)
|
|
|
|
p = write_position_timeline(run_dir, trades, open_positions)
|
|
if p:
|
|
paths["position_timeline"] = str(p)
|
|
|
|
# Plots (no-op)
|
|
paths["plots_dir"] = str(write_plots(run_dir, config.reporting.generate_plots))
|
|
|
|
# Notes
|
|
paths["run_notes"] = str(write_run_notes(run_dir, manifest.notes or ""))
|
|
|
|
return paths
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Internal helpers
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
def _write_parquet(rows: list[dict[str, Any]], path: Path) -> None:
|
|
"""Write list of row dicts to Parquet."""
|
|
if not rows:
|
|
return
|
|
keys = list(rows[0].keys())
|
|
columns: dict[str, list[Any]] = {k: [] for k in keys}
|
|
for row in rows:
|
|
for k in keys:
|
|
columns[k].append(row.get(k))
|
|
table = pa.table({k: pa.array(v) for k, v in columns.items()})
|
|
pq.write_table(table, str(path))
|
|
logger.debug("parquet_written", path=str(path), rows=len(rows))
|