File size: 9,217 Bytes
a47b09f | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 | """
Main benchmark runner: orchestrates data generation, algorithm runs, feature extraction,
and result collection into a meta-dataset.
"""
import os
import json
import time
import numpy as np
import pandas as pd
import logging
import warnings
from datetime import datetime
from causal_selection.data.generator import (
load_bn_model, get_true_dag_adjmat, dag_to_cpdag, sample_dataset,
SMALL_NETWORKS, MEDIUM_NETWORKS, LARGE_NETWORKS, ALL_NETWORKS,
SAMPLE_SIZES, SEEDS_PER_CONFIG, 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
warnings.filterwarnings('ignore')
logging.basicConfig(level=logging.INFO, format='%(asctime)s %(levelname)s %(message)s')
logger = logging.getLogger(__name__)
RESULTS_DIR = '/app/causal_selection/data/results'
ALGO_NAMES = list(ALGORITHM_POOL.keys())
# Timeout per algorithm per dataset (seconds)
TIMEOUT_MAP = {
'small': 60, # 1 min for small networks
'medium': 180, # 3 min for medium networks
'large': 300, # 5 min for large networks
}
def run_single_config(network, n_samples, seed, timeout_sec=300):
"""Run all algorithms on a single (network, n_samples, seed) configuration.
Returns:
dict with:
- 'meta_features': dict of feature values
- 'metrics': dict of algo_name -> metrics dict
- 'config': dict with network, n_samples, seed
"""
logger.info(f"=== {network} N={n_samples} seed={seed} ===")
# Load network and ground truth
model = load_bn_model(network)
true_dag, node_names = get_true_dag_adjmat(model)
true_cpdag = dag_to_cpdag(true_dag)
# Sample data
t0 = time.time()
df = sample_dataset(model, n_samples, seed=seed)
sample_time = time.time() - t0
logger.info(f" Sampled {df.shape} in {sample_time:.1f}s")
# Extract meta-features
t0 = time.time()
features = extract_all_features(df, n_probe_triplets=100)
feat_time = time.time() - t0
logger.info(f" Extracted {len(features)} features in {feat_time:.1f}s")
# Run all algorithms
algo_metrics = {}
for algo_name in ALGO_NAMES:
t0 = time.time()
result = run_algorithm(algo_name, df, timeout_sec=timeout_sec)
metrics = evaluate_algorithm_result(result, true_cpdag)
algo_metrics[algo_name] = metrics
status_str = metrics['status']
if status_str == 'success':
logger.info(f" {algo_name:15s}: SHD={metrics['shd']:3d} F1={metrics['skeleton_f1']:.3f} "
f"time={metrics['runtime']:.1f}s")
else:
logger.info(f" {algo_name:15s}: {status_str} time={metrics['runtime']:.1f}s")
return {
'meta_features': features,
'metrics': algo_metrics,
'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),
}
}
def build_meta_dataset(networks=None, save_intermediate=True):
"""Run full benchmark and build meta-dataset.
Returns:
X: pd.DataFrame of meta-features
Y_shd: pd.DataFrame of SHD per algorithm (columns = algo names)
Y_nshd: pd.DataFrame of normalized SHD
configs: list of config dicts
full_results: list of full result dicts
"""
if networks is None:
networks = ALL_NETWORKS
all_features = []
all_shd = []
all_nshd = []
all_configs = []
full_results = []
total_configs = 0
for net in networks:
tier = get_network_tier(net)
n_sizes = len(SAMPLE_SIZES[tier])
total_configs += n_sizes * SEEDS_PER_CONFIG
logger.info(f"Starting benchmark: {len(networks)} networks, ~{total_configs} configs")
config_idx = 0
for network in networks:
tier = get_network_tier(network)
sample_sizes = SAMPLE_SIZES[tier]
timeout = TIMEOUT_MAP[tier]
for n_samples in sample_sizes:
for seed in range(SEEDS_PER_CONFIG):
config_idx += 1
logger.info(f"\n[{config_idx}/{total_configs}] "
f"{network} N={n_samples} seed={seed}")
try:
result = run_single_config(network, n_samples, seed,
timeout_sec=timeout)
# Extract feature vector
feat_row = {name: result['meta_features'].get(name, 0.0)
for name in FEATURE_NAMES}
all_features.append(feat_row)
# Extract SHD vector
shd_row = {}
nshd_row = {}
for algo in ALGO_NAMES:
m = result['metrics'][algo]
shd_row[algo] = m['shd']
nshd_row[algo] = m['normalized_shd']
all_shd.append(shd_row)
all_nshd.append(nshd_row)
# Config info
all_configs.append(result['config'])
full_results.append(result)
except Exception as e:
logger.error(f"FAILED config {network} N={n_samples} seed={seed}: {e}")
continue
# Save intermediate results periodically
if save_intermediate and config_idx % 5 == 0:
_save_intermediate(all_features, all_shd, all_nshd, all_configs)
# Build final DataFrames
X = pd.DataFrame(all_features, columns=FEATURE_NAMES)
Y_shd = pd.DataFrame(all_shd, columns=ALGO_NAMES)
Y_nshd = pd.DataFrame(all_nshd, columns=ALGO_NAMES)
configs_df = pd.DataFrame(all_configs)
# Save final results
os.makedirs(RESULTS_DIR, exist_ok=True)
X.to_csv(os.path.join(RESULTS_DIR, 'meta_features.csv'), index=False)
Y_shd.to_csv(os.path.join(RESULTS_DIR, 'shd_matrix.csv'), index=False)
Y_nshd.to_csv(os.path.join(RESULTS_DIR, 'normalized_shd_matrix.csv'), index=False)
configs_df.to_csv(os.path.join(RESULTS_DIR, 'configs.csv'), index=False)
# Save full results as JSON
_save_full_results(full_results)
logger.info(f"\n=== BENCHMARK COMPLETE ===")
logger.info(f"Total configs: {len(all_features)}")
logger.info(f"Meta-feature matrix: {X.shape}")
logger.info(f"SHD matrix: {Y_shd.shape}")
logger.info(f"Results saved to {RESULTS_DIR}")
return X, Y_shd, Y_nshd, configs_df, full_results
def _save_intermediate(features, shds, nshds, configs):
"""Save intermediate results."""
os.makedirs(RESULTS_DIR, exist_ok=True)
pd.DataFrame(features).to_csv(os.path.join(RESULTS_DIR, 'meta_features_partial.csv'), index=False)
pd.DataFrame(shds).to_csv(os.path.join(RESULTS_DIR, 'shd_matrix_partial.csv'), index=False)
pd.DataFrame(nshds).to_csv(os.path.join(RESULTS_DIR, 'normalized_shd_partial.csv'), index=False)
pd.DataFrame(configs).to_csv(os.path.join(RESULTS_DIR, 'configs_partial.csv'), index=False)
def _save_full_results(results):
"""Save full results (without numpy arrays)."""
serializable = []
for r in results:
entry = {
'config': r['config'],
'meta_features': {k: float(v) if isinstance(v, (np.floating, np.integer)) else v
for k, v in r['meta_features'].items()},
'metrics': {}
}
for algo, m in r['metrics'].items():
entry['metrics'][algo] = {
k: float(v) if isinstance(v, (np.floating, np.integer)) else v
for k, v in m.items()
}
serializable.append(entry)
with open(os.path.join(RESULTS_DIR, 'full_results.json'), 'w') as f:
json.dump(serializable, f, indent=2)
if __name__ == '__main__':
import sys
# Allow selecting network tier from command line
tier = sys.argv[1] if len(sys.argv) > 1 else 'small'
if tier == 'small':
networks = SMALL_NETWORKS
elif tier == 'medium':
networks = MEDIUM_NETWORKS
elif tier == 'large':
networks = LARGE_NETWORKS
elif tier == 'all':
networks = ALL_NETWORKS
else:
networks = [tier] # single network name
logger.info(f"Running benchmark for tier: {tier} ({networks})")
X, Y_shd, Y_nshd, configs, results = build_meta_dataset(networks=networks)
# Print summary
print("\n" + "=" * 80)
print("BENCHMARK SUMMARY")
print("=" * 80)
print(f"\nMeta-feature matrix: {X.shape}")
print(f"SHD matrix: {Y_shd.shape}")
print(f"\nMean SHD per algorithm:")
print(Y_shd.mean().sort_values().to_string())
print(f"\nBest algorithm per config:")
best = Y_shd.idxmin(axis=1)
print(best.value_counts().to_string())
|