File size: 7,967 Bytes
fbf08d1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 | """
Resume benchmark from partial results, then run medium and large networks too.
Optimized for CPU: reduced timeouts, skip heavy combos.
"""
import os
import sys
import time
import numpy as np
import pandas as pd
import json
import logging
import warnings
warnings.filterwarnings('ignore')
logging.basicConfig(level=logging.INFO, format='%(asctime)s %(levelname)s %(message)s')
logger = logging.getLogger(__name__)
# Suppress the verbose BOSS/GRaSP output
logging.getLogger('causallearn').setLevel(logging.WARNING)
sys.path.insert(0, '/app')
from causal_selection.data.generator import (
load_bn_model, get_true_dag_adjmat, dag_to_cpdag, sample_dataset,
SMALL_NETWORKS, MEDIUM_NETWORKS, LARGE_NETWORKS, get_network_tier
)
from causal_selection.discovery.algorithms import run_algorithm, ALGORITHM_POOL
from causal_selection.discovery.evaluator import evaluate_algorithm_result
from causal_selection.features.extractor import extract_all_features, FEATURE_NAMES
ALGO_NAMES = list(ALGORITHM_POOL.keys())
RESULTS_DIR = '/app/causal_selection/data/results'
# More aggressive sample sizes - fewer but covering range
SAMPLE_SIZES_FAST = {
'small': [500, 1000, 2000, 5000],
'medium': [500, 1000, 2000],
'large': [500, 1000],
}
# Per-algorithm timeout (seconds) - algorithm-specific!
ALGO_TIMEOUT = {
'PC_discrete': {'small': 30, 'medium': 120, 'large': 300},
'FCI': {'small': 30, 'medium': 120, 'large': 300},
'GES': {'small': 30, 'medium': 120, 'large': 300},
'BOSS': {'small': 30, 'medium': 120, 'large': 300},
'GRaSP': {'small': 30, 'medium': 120, 'large': 300},
'HC': {'small': 30, 'medium': 60, 'large': 120},
'Tabu': {'small': 30, 'medium': 60, 'large': 120},
'MMHC': {'small': 30, 'medium': 60, 'large': 120},
'K2': {'small': 20, 'medium': 30, 'large': 60},
}
SEEDS = 2 # Reduced from 3 to speed up
def load_existing_results():
"""Load existing partial results to avoid re-running."""
existing = set()
partial_path = os.path.join(RESULTS_DIR, 'configs_partial.csv')
final_path = os.path.join(RESULTS_DIR, 'configs.csv')
for path in [partial_path, final_path]:
if os.path.exists(path):
df = pd.read_csv(path)
for _, row in df.iterrows():
key = (row['network'], int(row['n_samples']), int(row['seed']))
existing.add(key)
return existing
def run_benchmark():
"""Run full benchmark with resume capability."""
existing = load_existing_results()
logger.info(f"Found {len(existing)} existing configs")
# Load existing partial data
all_features = []
all_shd = []
all_nshd = []
all_configs = []
for prefix in ['meta_features_partial', 'meta_features']:
path = os.path.join(RESULTS_DIR, f'{prefix}.csv')
if os.path.exists(path):
df = pd.read_csv(path)
all_features = df.to_dict('records')
break
for prefix in ['shd_matrix_partial', 'shd_matrix']:
path = os.path.join(RESULTS_DIR, f'{prefix}.csv')
if os.path.exists(path):
df = pd.read_csv(path)
all_shd = df.to_dict('records')
break
for prefix in ['normalized_shd_partial', 'normalized_shd_matrix']:
path = os.path.join(RESULTS_DIR, f'{prefix}.csv')
if os.path.exists(path):
df = pd.read_csv(path)
all_nshd = df.to_dict('records')
break
for prefix in ['configs_partial', 'configs']:
path = os.path.join(RESULTS_DIR, f'{prefix}.csv')
if os.path.exists(path):
df = pd.read_csv(path)
all_configs = df.to_dict('records')
break
logger.info(f"Starting with {len(all_configs)} existing results")
# Generate all configs to run
all_networks = SMALL_NETWORKS + MEDIUM_NETWORKS + LARGE_NETWORKS
configs_to_run = []
for net in all_networks:
tier = get_network_tier(net)
for n_samples in SAMPLE_SIZES_FAST[tier]:
for seed in range(SEEDS):
key = (net, n_samples, seed)
if key not in existing:
configs_to_run.append((net, n_samples, seed, tier))
logger.info(f"Configs to run: {len(configs_to_run)}")
total = len(configs_to_run)
for idx, (network, n_samples, seed, tier) in enumerate(configs_to_run):
logger.info(f"\n[{idx+1}/{total}] {network} N={n_samples} seed={seed}")
try:
# Load network
model = load_bn_model(network)
true_dag, node_names = get_true_dag_adjmat(model)
true_cpdag = dag_to_cpdag(true_dag)
# Sample data
df = sample_dataset(model, n_samples, seed=seed)
# Extract features
features = extract_all_features(df, n_probe_triplets=80)
# Run algorithms with per-algo timeout
algo_metrics = {}
for algo_name in ALGO_NAMES:
timeout = ALGO_TIMEOUT[algo_name][tier]
result = run_algorithm(algo_name, df, timeout_sec=timeout)
metrics = evaluate_algorithm_result(result, true_cpdag)
algo_metrics[algo_name] = metrics
s = metrics['status']
if s == 'success':
logger.info(f" {algo_name:12s}: SHD={metrics['shd']:3d} F1={metrics['skeleton_f1']:.3f} t={metrics['runtime']:.1f}s")
else:
logger.info(f" {algo_name:12s}: {s} t={metrics['runtime']:.1f}s")
# Store results
feat_row = {name: features.get(name, 0.0) for name in FEATURE_NAMES}
all_features.append(feat_row)
shd_row = {algo: algo_metrics[algo]['shd'] for algo in ALGO_NAMES}
nshd_row = {algo: algo_metrics[algo]['normalized_shd'] for algo in ALGO_NAMES}
all_shd.append(shd_row)
all_nshd.append(nshd_row)
config = {
'network': network,
'n_samples': n_samples,
'seed': seed,
'n_variables': len(node_names),
'n_true_edges': int(((true_cpdag + true_cpdag.T) > 0).sum() // 2),
}
all_configs.append(config)
# Save periodically
if (idx + 1) % 3 == 0:
_save_results(all_features, all_shd, all_nshd, all_configs, partial=True)
except Exception as e:
logger.error(f"FAILED {network} N={n_samples} seed={seed}: {e}")
import traceback
traceback.print_exc()
# Final save
_save_results(all_features, all_shd, all_nshd, all_configs, partial=False)
# Print summary
Y_shd = pd.DataFrame(all_shd)
configs_df = pd.DataFrame(all_configs)
print("\n" + "=" * 80)
print("BENCHMARK COMPLETE")
print("=" * 80)
print(f"Total configs: {len(all_configs)}")
print(f"Networks: {configs_df['network'].unique()}")
print(f"\nMean SHD per algorithm:")
print(Y_shd.mean().sort_values())
print(f"\nBest algorithm count:")
print(Y_shd.idxmin(axis=1).value_counts())
def _save_results(features, shds, nshds, configs, partial=True):
os.makedirs(RESULTS_DIR, exist_ok=True)
suffix = '_partial' if partial else ''
pd.DataFrame(features).to_csv(os.path.join(RESULTS_DIR, f'meta_features{suffix}.csv'), index=False)
pd.DataFrame(shds).to_csv(os.path.join(RESULTS_DIR, f'shd_matrix{suffix}.csv'), index=False)
pd.DataFrame(nshds).to_csv(os.path.join(RESULTS_DIR, f'normalized_shd_matrix{suffix}.csv'), index=False)
pd.DataFrame(configs).to_csv(os.path.join(RESULTS_DIR, f'configs{suffix}.csv'), index=False)
if __name__ == '__main__':
run_benchmark()
|