ClaudeandClaude Opus 4.6
7bc4aac0aa
The port of run_batch_subprocess from lab_api_backup.py to services/forge_runner.py truncated the function signature to 8 params + **kwargs, but the route handler in routes/lab.py passes 13 positional arguments, causing TypeError at runtime when POST /api/lab/start fires. Restore all 13 explicit parameters (policy_style, policy_greedy, ai_simplified, ai_think_time_ms, max_queue_depth) to match the caller and pass them as keyword args to build_java_command. Also compute policy_server from CFG.port and add error handling, matching the original backup implementation. Additionally, _run_process_blocking was writing to wrong BatchState attribute names (state.status, state.games_done, state.games_total, state.throughput, state.log, state.started_at, state.ended_at) instead of the correct ones (state.running, state.completed_games, state.total_games, state.sims_per_sec, state.log_lines, state.elapsed_ms). This caused the /api/lab/status endpoint to never see progress updates. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
483 lines
20 KiB
Python
483 lines
20 KiB
Python
"""Forge batch simulation runner: Java discovery, command building, process management."""
|
|
import asyncio
|
|
import json
|
|
import logging
|
|
import os
|
|
import re
|
|
import subprocess
|
|
import sys
|
|
import threading
|
|
import time
|
|
from datetime import datetime as _datetime
|
|
from pathlib import Path
|
|
from typing import Optional
|
|
|
|
from models.state import CFG, BatchState
|
|
from services.deck_service import parse_dck_file, _load_deck_cards_by_name
|
|
|
|
log = logging.getLogger("lab.batch")
|
|
log_sim = logging.getLogger("commander_ai_lab.sim")
|
|
|
|
# Global DeepSeek brain instance (lazy-initialized)
|
|
_deepseek_brain = None
|
|
_deepseek_lock = threading.Lock()
|
|
|
|
|
|
def _find_java17() -> str:
|
|
search_dirs = [r'C:\Program Files\Eclipse Adoptium', r'C:\Program Files\Java']
|
|
for d in search_dirs:
|
|
if os.path.isdir(d):
|
|
for child in os.listdir(d):
|
|
if child.startswith('jdk-17'):
|
|
candidate = os.path.join(d, child, 'bin', 'java.exe')
|
|
if os.path.isfile(candidate):
|
|
return candidate
|
|
return 'java'
|
|
|
|
|
|
_JAVA17_PATH = None
|
|
|
|
|
|
def get_java17() -> str:
|
|
global _JAVA17_PATH
|
|
if _JAVA17_PATH is None:
|
|
_JAVA17_PATH = _find_java17()
|
|
return _JAVA17_PATH
|
|
|
|
|
|
def build_java_command(
|
|
decks: list, num_games: int, threads: int, seed: Optional[int],
|
|
clock: int, output_path: str, use_learned_policy: bool = False,
|
|
policy_server: str = "http://localhost:8080", policy_style: str = "midrange",
|
|
policy_greedy: bool = False, ai_simplified: bool = False,
|
|
ai_think_time_ms: int = -1, max_queue_depth: int = -1,
|
|
) -> list:
|
|
java17 = get_java17()
|
|
forge_jar = CFG.forge_jar
|
|
cmd = [
|
|
java17, '-jar', forge_jar, '-sim',
|
|
'-decks', ','.join(str(d) for d in decks),
|
|
'-games', str(num_games),
|
|
'-threads', str(threads),
|
|
'-clock', str(clock),
|
|
'-output', output_path,
|
|
]
|
|
if seed is not None:
|
|
cmd += ['-seed', str(seed)]
|
|
if use_learned_policy:
|
|
cmd += ['-policy', policy_style, '-policyServer', policy_server]
|
|
if policy_greedy:
|
|
cmd += ['-policyGreedy']
|
|
if ai_simplified:
|
|
cmd += ['-aiSimplified']
|
|
if ai_think_time_ms > 0:
|
|
cmd += ['-aiThinkTimeMs', str(ai_think_time_ms)]
|
|
if max_queue_depth > 0:
|
|
cmd += ['-maxQueueDepth', str(max_queue_depth)]
|
|
return cmd
|
|
|
|
|
|
def _run_process_blocking(state: BatchState, cmd: list):
|
|
"""Run Forge subprocess, parse progress, generate deck reports."""
|
|
log.info(f"Running: {' '.join(cmd)}")
|
|
|
|
# Ensure Forge subprocesses use Java 17
|
|
env = os.environ.copy()
|
|
java17 = get_java17()
|
|
if java17 != 'java':
|
|
java17_bin = os.path.dirname(java17)
|
|
java17_home = os.path.dirname(java17_bin)
|
|
env['JAVA_HOME'] = java17_home
|
|
env['PATH'] = java17_bin + os.pathsep + env.get('PATH', '')
|
|
|
|
try:
|
|
proc = subprocess.Popen(
|
|
cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT,
|
|
text=True, bufsize=1, env=env,
|
|
)
|
|
state.process = proc
|
|
except Exception as e:
|
|
state.running = False
|
|
state.error = str(e)
|
|
log.error(f"Failed to start subprocess: {e}")
|
|
return
|
|
|
|
last_activity = [time.monotonic()]
|
|
stall_limit = 300 # 5 min
|
|
|
|
def watchdog():
|
|
while proc.poll() is None:
|
|
if (time.monotonic() - last_activity[0]) > stall_limit:
|
|
log.warning("Stall detected, killing process")
|
|
state.log_lines.append('[WATCHDOG] Process stalled. Killed.')
|
|
proc.kill()
|
|
state.running = False
|
|
state.error = "Stall detected (5 min no output)"
|
|
return
|
|
time.sleep(5)
|
|
|
|
wd = threading.Thread(target=watchdog, daemon=True)
|
|
wd.start()
|
|
|
|
for line in proc.stdout:
|
|
last_activity[0] = time.monotonic()
|
|
line = line.rstrip()
|
|
state.log_lines.append(line)
|
|
gm = re.match(r'\[Game (\d+)/(\d+)\]', line)
|
|
if gm:
|
|
state.completed_games = int(gm.group(1))
|
|
state.total_games = int(gm.group(2))
|
|
pm = re.match(r'\[PROGRESS\].*?(\d+\.?\d*)\s*sims/sec', line)
|
|
if pm:
|
|
state.sims_per_sec = float(pm.group(1))
|
|
|
|
proc.wait()
|
|
elapsed = (_datetime.now() - state.start_time).total_seconds() * 1000
|
|
state.elapsed_ms = int(elapsed)
|
|
if proc.returncode != 0:
|
|
state.running = False
|
|
state.error = f"Exit code {proc.returncode}"
|
|
else:
|
|
state.running = False
|
|
state.completed_games = state.total_games
|
|
|
|
|
|
async def run_batch_subprocess(
|
|
state: BatchState,
|
|
decks: list,
|
|
num_games: int,
|
|
threads: int,
|
|
seed: Optional[int] = None,
|
|
clock: int = 6000,
|
|
output_path: str = "results",
|
|
use_learned_policy: bool = False,
|
|
policy_style: str = "midrange",
|
|
policy_greedy: bool = False,
|
|
ai_simplified: bool = False,
|
|
ai_think_time_ms: int = -1,
|
|
max_queue_depth: int = -1,
|
|
):
|
|
"""Async wrapper that runs Forge subprocess in executor."""
|
|
try:
|
|
policy_server = f"http://localhost:{CFG.port}"
|
|
cmd = build_java_command(
|
|
decks, num_games, threads, seed, clock, output_path,
|
|
use_learned_policy=use_learned_policy,
|
|
policy_server=policy_server,
|
|
policy_style=policy_style,
|
|
policy_greedy=policy_greedy,
|
|
ai_simplified=ai_simplified,
|
|
ai_think_time_ms=ai_think_time_ms,
|
|
max_queue_depth=max_queue_depth,
|
|
)
|
|
log.info(f"Starting batch {state.batch_id}: {' '.join(cmd[:6])}...")
|
|
loop = asyncio.get_event_loop()
|
|
await loop.run_in_executor(None, _run_process_blocking, state, cmd)
|
|
except Exception as e:
|
|
state.error = str(e)
|
|
state.running = False
|
|
log.error(f"Batch {state.batch_id} ERROR: {e}")
|
|
|
|
|
|
def _get_deepseek_brain():
|
|
"""Get or create the global DeepSeek brain instance."""
|
|
global _deepseek_brain
|
|
if _deepseek_brain is None:
|
|
with _deepseek_lock:
|
|
if _deepseek_brain is None:
|
|
try:
|
|
src_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', 'src')
|
|
src_dir = os.path.normpath(src_dir)
|
|
if src_dir not in sys.path:
|
|
sys.path.insert(0, src_dir)
|
|
from commander_ai_lab.sim.deepseek_brain import DeepSeekBrain, DeepSeekConfig
|
|
cfg = DeepSeekConfig()
|
|
# Allow env var overrides
|
|
if os.environ.get('DEEPSEEK_API_BASE'):
|
|
cfg.api_base = os.environ['DEEPSEEK_API_BASE']
|
|
if os.environ.get('DEEPSEEK_MODEL'):
|
|
cfg.model = os.environ['DEEPSEEK_MODEL']
|
|
cfg.log_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', 'logs', 'decisions')
|
|
cfg.log_dir = os.path.normpath(cfg.log_dir)
|
|
_deepseek_brain = DeepSeekBrain(cfg)
|
|
except Exception as e:
|
|
log_sim.error(f'Failed to initialize brain: {e}')
|
|
return None
|
|
return _deepseek_brain
|
|
|
|
|
|
def _run_deepseek_batch_thread(
|
|
state: BatchState,
|
|
deck_names: list,
|
|
num_games: int,
|
|
output_path: str,
|
|
):
|
|
"""Run batch simulation using Python sim engine + DeepSeek AI opponent."""
|
|
try:
|
|
src_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', 'src')
|
|
src_dir = os.path.normpath(src_dir)
|
|
if src_dir not in sys.path:
|
|
sys.path.insert(0, src_dir)
|
|
|
|
from commander_ai_lab.sim.models import Card
|
|
from commander_ai_lab.sim.deepseek_engine import DeepSeekGameEngine
|
|
from commander_ai_lab.sim.rules import enrich_card
|
|
|
|
state.log_lines.append('[DeepSeek Batch] Initializing DeepSeek brain...')
|
|
brain = _get_deepseek_brain()
|
|
if brain and not brain._connected:
|
|
brain.check_connection()
|
|
if not brain or not brain._connected:
|
|
state.error = 'DeepSeek LLM not connected. Go to Simulator > DeepSeek and connect first.'
|
|
state.running = False
|
|
return
|
|
|
|
state.log_lines.append(f'[DeepSeek Batch] Connected to {brain.config.model}')
|
|
|
|
# Load all decks
|
|
# ── Look up historical win rates from past batch results ──
|
|
deck_win_rates = {} # deck_name -> float (0-100)
|
|
try:
|
|
results_dir = CFG.results_dir
|
|
if os.path.isdir(results_dir):
|
|
for fname in sorted(os.listdir(results_dir), reverse=True):
|
|
if fname.startswith('batch-') and fname.endswith('.json'):
|
|
try:
|
|
with open(os.path.join(results_dir, fname), 'r') as rf:
|
|
past = json.loads(rf.read())
|
|
for dd in past.get('decks', []):
|
|
dname = dd.get('deckName', '')
|
|
wr = dd.get('winRate')
|
|
if dname and wr is not None and dname not in deck_win_rates:
|
|
deck_win_rates[dname] = float(wr)
|
|
except Exception:
|
|
pass
|
|
except Exception:
|
|
pass
|
|
|
|
loaded_decks = {}
|
|
deck_meta = {} # deck_name -> {commander_name, color_identity, archetype}
|
|
for dn in deck_names:
|
|
raw_cards = _load_deck_cards_by_name(dn)
|
|
if not raw_cards:
|
|
state.log_lines.append(f'[DeepSeek Batch] WARNING: Could not load deck "{dn}", skipping.')
|
|
continue
|
|
|
|
# Detect commander and color identity from card data
|
|
commander_name = ''
|
|
color_identity_set = set()
|
|
deck_objs = []
|
|
for cd in raw_cards:
|
|
c = Card(name=cd['name'])
|
|
if cd.get('type_line'): c.type_line = cd['type_line']
|
|
if cd.get('cmc'): c.cmc = float(cd['cmc'])
|
|
if cd.get('power') and cd.get('toughness'):
|
|
c.power = str(cd['power'])
|
|
c.toughness = str(cd['toughness'])
|
|
c.pt = c.power + '/' + c.toughness
|
|
if cd.get('oracle_text'): c.oracle_text = cd['oracle_text']
|
|
if cd.get('mana_cost'): c.mana_cost = cd['mana_cost']
|
|
if cd.get('keywords'):
|
|
kw = cd['keywords']
|
|
if isinstance(kw, str):
|
|
try: kw = json.loads(kw)
|
|
except Exception: kw = []
|
|
if isinstance(kw, list): c.keywords = kw
|
|
# Set is_commander flag from DB data
|
|
if cd.get('is_commander'):
|
|
c.is_commander = True
|
|
commander_name = cd['name']
|
|
# Collect color identity
|
|
ci_str = cd.get('color_identity', '')
|
|
if ci_str:
|
|
try:
|
|
ci_parsed = json.loads(ci_str) if isinstance(ci_str, str) else ci_str
|
|
if isinstance(ci_parsed, list):
|
|
for color in ci_parsed:
|
|
color_identity_set.add(color)
|
|
except Exception:
|
|
pass
|
|
enrich_card(c)
|
|
deck_objs.append(c)
|
|
|
|
# Infer archetype from deck composition
|
|
creature_count = sum(1 for c in deck_objs if c.is_creature())
|
|
removal_count = sum(1 for c in deck_objs if c.is_removal)
|
|
ramp_count = sum(1 for c in deck_objs if c.is_ramp)
|
|
avg_cmc = sum(c.cmc or 0 for c in deck_objs if not c.is_land()) / max(sum(1 for c in deck_objs if not c.is_land()), 1)
|
|
oracle_all = ' '.join((c.oracle_text or '').lower() for c in deck_objs)
|
|
has_combo_text = any(kw in oracle_all for kw in ['you win the game', 'infinite', 'extra turn'])
|
|
|
|
if has_combo_text:
|
|
archetype = 'combo'
|
|
elif creature_count >= 30 and avg_cmc <= 2.8:
|
|
archetype = 'aggro'
|
|
elif removal_count >= 10 or (creature_count <= 18 and avg_cmc >= 3.2):
|
|
archetype = 'control'
|
|
else:
|
|
archetype = 'midrange'
|
|
|
|
color_identity_list = sorted(list(color_identity_set))
|
|
deck_meta[dn] = {
|
|
'commander_name': commander_name,
|
|
'color_identity': color_identity_list,
|
|
'archetype': archetype,
|
|
'win_rate': deck_win_rates.get(dn),
|
|
}
|
|
|
|
loaded_decks[dn] = deck_objs
|
|
cmdr_info = f' (Commander: {commander_name})' if commander_name else ''
|
|
wr_info = f' [History: {deck_win_rates[dn]:.0f}% WR]' if dn in deck_win_rates else ''
|
|
state.log_lines.append(f'[DeepSeek Batch] Loaded deck "{dn}" ({len(deck_objs)} cards, {archetype}){cmdr_info}{wr_info}')
|
|
|
|
if not loaded_decks:
|
|
state.error = 'No decks could be loaded.'
|
|
state.running = False
|
|
return
|
|
|
|
# Build matchup schedule: each deck plays num_games vs DeepSeek AI
|
|
deck_list = list(loaded_decks.keys())
|
|
games_per_deck = max(1, num_games // len(deck_list))
|
|
total_games = games_per_deck * len(deck_list)
|
|
state.total_games = total_games
|
|
|
|
state.log_lines.append(f'[DeepSeek Batch] Running {games_per_deck} games per deck \u00d7 {len(deck_list)} decks = {total_games} total')
|
|
|
|
engine = DeepSeekGameEngine(
|
|
brain=brain,
|
|
ai_player_index=0, # AI pilots deck_a (user's deck) with full intelligence
|
|
max_turns=25,
|
|
record_log=True,
|
|
ml_log=True,
|
|
)
|
|
|
|
start_time = time.time()
|
|
all_deck_results = []
|
|
completed = 0
|
|
|
|
for deck_name in deck_list:
|
|
deck_a = loaded_decks[deck_name]
|
|
# Generate a training opponent for each game
|
|
from commander_ai_lab.lab.experiments import _generate_training_deck
|
|
|
|
deck_stats = {
|
|
'deckName': deck_name,
|
|
'wins': 0, 'losses': 0, 'totalGames': games_per_deck,
|
|
'totalTurns': 0, 'totalDamageDealt': 0, 'totalDamageReceived': 0,
|
|
'totalSpellsCast': 0, 'totalCreaturesPlayed': 0,
|
|
'games': [],
|
|
}
|
|
|
|
meta = deck_meta.get(deck_name, {})
|
|
dk_archetype = meta.get('archetype', 'midrange')
|
|
dk_commander = meta.get('commander_name', '')
|
|
dk_colors = meta.get('color_identity', [])
|
|
dk_win_rate = meta.get('win_rate')
|
|
|
|
for g in range(games_per_deck):
|
|
try:
|
|
deck_b = _generate_training_deck()
|
|
game_id = f'ds-{state.batch_id}-{deck_name[:12]}-g{g+1}'
|
|
result = engine.run(
|
|
deck_a, deck_b,
|
|
name_a=deck_name + ' (AI)',
|
|
name_b='Training Opponent',
|
|
game_id=game_id, archetype=dk_archetype,
|
|
commander_name=dk_commander,
|
|
color_identity=dk_colors,
|
|
win_rate=dk_win_rate,
|
|
)
|
|
gd = result.to_dict()
|
|
gd['gameNumber'] = g + 1
|
|
deck_stats['games'].append(gd)
|
|
|
|
if result.winner == 0:
|
|
deck_stats['wins'] += 1
|
|
else:
|
|
deck_stats['losses'] += 1
|
|
deck_stats['totalTurns'] += result.turns
|
|
if result.player_a_stats:
|
|
deck_stats['totalDamageDealt'] += result.player_a_stats.damage_dealt
|
|
deck_stats['totalDamageReceived'] += result.player_a_stats.damage_received
|
|
deck_stats['totalSpellsCast'] += result.player_a_stats.spells_cast
|
|
deck_stats['totalCreaturesPlayed'] += result.player_a_stats.creatures_played
|
|
|
|
state.log_lines.append(
|
|
f'[Game {completed + 1}/{total_games}] {deck_name} (AI-piloted) \u2192 '
|
|
f'{"WIN" if result.winner == 0 else "LOSS"} (turn {result.turns})'
|
|
)
|
|
except Exception as ge:
|
|
state.log_lines.append(f'[Game {completed + 1}/{total_games}] ERROR: {ge}')
|
|
deck_stats['games'].append({'error': str(ge), 'gameNumber': g + 1})
|
|
|
|
completed += 1
|
|
state.completed_games = completed
|
|
|
|
n = deck_stats['totalGames']
|
|
deck_stats['winRate'] = round(deck_stats['wins'] / n * 100, 1) if n > 0 else 0.0
|
|
deck_stats['avgTurns'] = round(deck_stats['totalTurns'] / n, 1) if n > 0 else 0.0
|
|
deck_stats['avgDamageDealt'] = round(deck_stats['totalDamageDealt'] / n, 1) if n > 0 else 0.0
|
|
deck_stats['avgDamageReceived'] = round(deck_stats['totalDamageReceived'] / n, 1) if n > 0 else 0.0
|
|
deck_stats['avgSpellsCast'] = round(deck_stats['totalSpellsCast'] / n, 1) if n > 0 else 0.0
|
|
deck_stats['avgCreaturesPlayed'] = round(deck_stats['totalCreaturesPlayed'] / n, 1) if n > 0 else 0.0
|
|
# Include deck intelligence metadata
|
|
deck_stats['archetype'] = dk_archetype
|
|
deck_stats['commander'] = dk_commander
|
|
deck_stats['colorIdentity'] = dk_colors
|
|
if dk_win_rate is not None:
|
|
deck_stats['priorWinRate'] = dk_win_rate
|
|
all_deck_results.append(deck_stats)
|
|
|
|
elapsed = time.time() - start_time
|
|
ds_stats = brain.get_stats() if brain else {}
|
|
|
|
# Build result in compatible format
|
|
batch_result = {
|
|
'metadata': {
|
|
'batchId': state.batch_id,
|
|
'timestamp': _datetime.now().isoformat(),
|
|
'completedGames': completed,
|
|
'threads': 1,
|
|
'elapsedMs': int(elapsed * 1000),
|
|
'engine': 'deepseek',
|
|
'model': brain.config.model if brain else 'unknown',
|
|
},
|
|
'decks': all_deck_results,
|
|
'deepseekStats': ds_stats,
|
|
}
|
|
|
|
# Write ML decision JSONL for the training pipeline
|
|
ml_decisions = engine.flush_ml_decisions()
|
|
if ml_decisions:
|
|
ml_jsonl_path = os.path.join(CFG.results_dir, f'ml-decisions-ds-{state.batch_id}.jsonl')
|
|
os.makedirs(os.path.dirname(ml_jsonl_path) or '.', exist_ok=True)
|
|
with open(ml_jsonl_path, 'w', encoding='utf-8') as mf:
|
|
for dec in ml_decisions:
|
|
mf.write(json.dumps(dec) + '\n')
|
|
state.log_lines.append(f'[ML Data] Wrote {len(ml_decisions)} decision snapshots to {os.path.basename(ml_jsonl_path)}')
|
|
else:
|
|
state.log_lines.append('[ML Data] No decision snapshots captured (0 decisions)')
|
|
|
|
# Save to results dir
|
|
os.makedirs(os.path.dirname(output_path), exist_ok=True)
|
|
with open(output_path, 'w', encoding='utf-8') as f:
|
|
json.dump(batch_result, f, indent=2, default=str)
|
|
|
|
state.result_path = output_path
|
|
elapsed_ms = int(elapsed * 1000)
|
|
state.elapsed_ms = elapsed_ms
|
|
state.running = False
|
|
state.completed_games = total_games
|
|
|
|
state.log_lines.append(f'[DeepSeek Batch] Complete: {completed} games in {elapsed:.1f}s')
|
|
for ds in all_deck_results:
|
|
state.log_lines.append(f' {ds["deckName"]}: {ds["wins"]}W-{ds["losses"]}L ({ds["winRate"]}% WR)')
|
|
|
|
log_sim.info(f'Batch {state.batch_id} complete: {completed} games in {elapsed:.1f}s')
|
|
|
|
except Exception as e:
|
|
import traceback
|
|
state.error = str(e)
|
|
state.running = False
|
|
state.log_lines.append(f'[DeepSeek Batch] FATAL: {e}')
|
|
traceback.print_exc() |