add script to analyse dataset

This commit is contained in:
2026-04-03 15:45:54 +02:00
parent 49f2e0b008
commit 8e733dfe39
3 changed files with 268 additions and 0 deletions
+18
View File
@@ -133,6 +133,24 @@ just curate-dataset append=true
just curate-dataset append=true archive=true archive_dir=data/dataset/archive
```
Analyze dataset quality overall and by day (best game overall/day included):
```sh
python -m server.DatasetStats --input "good_moves-*.jsonl"
python -m server.DatasetStats --input data/dataset --output data/dataset/stats-report.json
```
The stats report now includes both:
- `best_game` (survival/length focused)
- `best_pressure_game` (high-pressure quality focused: fewer safe options + strong survival)
Or with `just`:
```sh
just analyze-dataset
just analyze-dataset input=data/dataset output=data/dataset/stats-report.json
```
To store compact dataset-only records (JSONL) and skip full per-game JSON files:
```sh
+3
View File
@@ -55,3 +55,6 @@ export-dataset input="data" output="data/dataset/good_moves.jsonl":
curate-dataset input="good_moves-*.jsonl" output="data/dataset/best_moves.jsonl" min_turn="6" late_turn="20" max_safe_options="2" min_score="3" append="false" archive="false" archive_dir="":
FLAGS=""; if [ "{{append}}" = "true" ]; then FLAGS="$FLAGS --append"; fi; if [ "{{archive}}" = "true" ]; then FLAGS="$FLAGS --archive-input"; fi; if [ -n "{{archive_dir}}" ]; then FLAGS="$FLAGS --archive-dir {{archive_dir}}"; fi; python -m server.DatasetCurator --input "{{input}}" --output "{{output}}" --min-turn "{{min_turn}}" --late-turn "{{late_turn}}" --max-safe-options "{{max_safe_options}}" --min-score "{{min_score}}" $FLAGS
analyze-dataset input="good_moves-*.jsonl" output="":
if [ -n "{{output}}" ]; then python -m server.DatasetStats --input "{{input}}" --output "{{output}}"; else python -m server.DatasetStats --input "{{input}}"; fi
+247
View File
@@ -0,0 +1,247 @@
import argparse
import glob
import json
import re
from collections import Counter, defaultdict
from datetime import datetime
from pathlib import Path
class DatasetStats:
DAY_PATTERN = re.compile(r"(\d{4}-\d{2}-\d{2})")
def __init__(self, input_files: list[str]):
self.input_files = input_files
def _resolve_input_files(self):
resolved = []
seen = set()
for item in self.input_files:
path = Path(item)
if path.is_dir():
for file_path in sorted(path.rglob("*.jsonl")):
key = str(file_path.resolve())
if key in seen:
continue
seen.add(key)
resolved.append(file_path)
continue
if any(ch in item for ch in "*?[]"):
for match in sorted(glob.glob(item)):
file_path = Path(match)
if not file_path.is_file():
continue
key = str(file_path.resolve())
if key in seen:
continue
seen.add(key)
resolved.append(file_path)
continue
if path.is_file():
key = str(path.resolve())
if key in seen:
continue
seen.add(key)
resolved.append(path)
return resolved
def _infer_day(self, file_path: Path):
match = self.DAY_PATTERN.search(file_path.name)
if match:
return match.group(1)
return datetime.fromtimestamp(file_path.stat().st_mtime).strftime("%Y-%m-%d")
def _game_score(self, game: dict):
max_turn = game["max_turn"]
rows = game["rows"]
avg_safe = game["avg_safe_options"]
pressure_bonus = 0 if avg_safe is None else max(0.0, 4.0 - avg_safe)
return round(max_turn * 2.0 + rows + pressure_bonus, 3)
def _pressure_score(self, game: dict):
max_turn = game["max_turn"]
rows = max(1, game["rows"])
pressure_turns = game["pressure_turns"]
avg_safe = game["avg_safe_options"]
pressure_ratio = pressure_turns / rows
safe_tightness = 0.0 if avg_safe is None else max(0.0, 3.0 - avg_safe)
return round(max_turn * 1.2 + pressure_ratio * 120.0 + safe_tightness * 20.0, 3)
def _extract_safe_options(self, row: dict):
top_level = row.get("safe_options")
if isinstance(top_level, int):
return top_level
history = row.get("history", {})
for item in history.get("data", []):
if item.get("function") != "get_possible_moves":
continue
safe_positions = item.get("safe_positions", {})
if isinstance(safe_positions, dict):
return len(safe_positions)
return None
def analyze(self):
files = self._resolve_input_files()
totals = {
"rows": 0,
"games": set(),
"snake_types": Counter(),
"game_types": Counter(),
"moves": Counter(),
"days": Counter(),
}
games = {}
day_games = defaultdict(set)
for file_path in files:
day = self._infer_day(file_path)
with file_path.open("r", encoding="utf-8") as source:
for line in source:
if not line.strip():
continue
row = json.loads(line)
game_id = row.get("game_id")
if not game_id:
continue
turn = int(row.get("turn", 0))
safe_options = self._extract_safe_options(row)
snake_type = row.get("snake_type", "unknown")
move = row.get("move", "unknown")
game_type = row.get("game_type", {})
if isinstance(game_type, dict):
game_type_name = game_type.get("name", "unknown")
else:
game_type_name = str(game_type)
totals["rows"] += 1
totals["games"].add(game_id)
totals["snake_types"][snake_type] += 1
totals["game_types"][game_type_name] += 1
totals["moves"][move] += 1
totals["days"][day] += 1
if game_id not in games:
games[game_id] = {
"game_id": game_id,
"day": day,
"snake_type": snake_type,
"game_type": game_type_name,
"rows": 0,
"max_turn": -1,
"safe_options_sum": 0,
"safe_options_count": 0,
"pressure_turns": 0,
}
game = games[game_id]
game["rows"] += 1
game["max_turn"] = max(game["max_turn"], turn)
if isinstance(safe_options, int):
game["safe_options_sum"] += safe_options
game["safe_options_count"] += 1
if safe_options <= 2:
game["pressure_turns"] += 1
day_games[day].add(game_id)
game_summaries = []
for game in games.values():
avg_safe = None
if game["safe_options_count"] > 0:
avg_safe = round(
game["safe_options_sum"] / game["safe_options_count"], 3
)
item = {
"game_id": game["game_id"],
"day": game["day"],
"snake_type": game["snake_type"],
"game_type": game["game_type"],
"rows": game["rows"],
"max_turn": game["max_turn"],
"avg_safe_options": avg_safe,
"pressure_turns": game["pressure_turns"],
}
item["score"] = self._game_score(item)
item["pressure_score"] = self._pressure_score(item)
game_summaries.append(item)
game_summaries.sort(
key=lambda x: (x["score"], x["max_turn"], x["rows"]), reverse=True
)
best_overall = game_summaries[0] if game_summaries else None
pressure_sorted = sorted(
game_summaries,
key=lambda x: (x["pressure_score"], x["max_turn"], x["rows"]),
reverse=True,
)
best_pressure_overall = pressure_sorted[0] if pressure_sorted else None
by_day = {}
for day, game_ids in sorted(day_games.items()):
day_list = [item for item in game_summaries if item["game_id"] in game_ids]
day_list.sort(
key=lambda x: (x["score"], x["max_turn"], x["rows"]), reverse=True
)
day_pressure = sorted(
day_list,
key=lambda x: (x["pressure_score"], x["max_turn"], x["rows"]),
reverse=True,
)
by_day[day] = {
"rows": totals["days"][day],
"games": len(game_ids),
"best_game": day_list[0] if day_list else None,
"best_pressure_game": day_pressure[0] if day_pressure else None,
}
return {
"files_scanned": [str(path) for path in files],
"overall": {
"rows": totals["rows"],
"games": len(totals["games"]),
"snake_types": dict(totals["snake_types"]),
"game_types": dict(totals["game_types"]),
"moves": dict(totals["moves"]),
"best_game": best_overall,
"best_pressure_game": best_pressure_overall,
},
"by_day": by_day,
"top_games": game_summaries[:10],
"top_pressure_games": pressure_sorted[:10],
}
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Analyze Battlesnake JSONL datasets")
parser.add_argument(
"--input",
action="append",
required=True,
help="Input JSONL file, directory, or glob pattern. Repeat for multiple inputs.",
)
parser.add_argument(
"--output",
default=None,
help="Optional path to write JSON report",
)
args = parser.parse_args()
report = DatasetStats(args.input).analyze()
print(json.dumps(report, indent=2))
if args.output:
output_path = Path(args.output)
output_path.parent.mkdir(parents=True, exist_ok=True)
output_path.write_text(json.dumps(report, indent=2), encoding="utf-8")