File size: 16,217 Bytes
1f8dd95 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 | #!/usr/bin/env python3
"""Run synthetic experiments for Dobrushin unlearning theory validation."""
import os
import sys
import json
import time
import argparse
import yaml
import numpy as np
from collections import defaultdict
from datetime import datetime
# Add project root to path
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from src.data import generate_gamma_poisson_data, generate_gaussian_gaussian_data, generate_gaussian_gamma_data, sample_deletions
from src.model import PoissonGammaVI, GaussianGaussianVI, GaussianGammaMAP, get_model
from src.graph_utils import build_adjacency, compute_graph_stats
from src.metrics import (compute_all_metrics, compute_deletion_influence_by_distance,
fit_exponential_decay, compute_local_error, compute_gradient_interference,
compute_chi_poisson_gamma, compute_chi_gaussian)
from src.unlearning import one_step_downdate_poisson_gamma
from src.utils import FitResult, generate_run_id, generate_config_id, save_jsonl, ensure_dir
def run_single_config(config, mode='pilot'):
"""Run all deletions for a single configuration."""
graph_type = config['graph_type']
N = config['N']
M = config['M']
K = config['K']
avg_degree = config['avg_degree']
count_scale = config.get('count_scale', 1.0)
prior_strength = config.get('prior_strength', 'strong')
prior_cfg = config.get('prior_config', {})
a0 = prior_cfg.get('a0', 0.3)
b0 = prior_cfg.get('b0', 1.0)
c0 = prior_cfg.get('c0', 0.3)
d0 = prior_cfg.get('d0', 1.0)
num_deletions = config.get('num_deletions', 20)
radii = config.get('radii', [1, 2, 3, 4])
max_iter = config.get('max_iter', 500)
tol = config.get('tol', 1e-5)
seed = config.get('seed', 42)
model_family = config.get('model_family', 'poisson_gamma')
config_id = generate_config_id(config)
run_id = generate_run_id()
print(f"\n{'='*60}")
print(f"Config: {graph_type}, K={K}, deg={avg_degree}, count={count_scale}, prior={prior_strength}")
print(f"Model: {model_family}, Config ID: {config_id}")
print(f"{'='*60}")
# Generate data
if model_family == 'poisson_gamma':
edges, U_true, V_true, graph_edges = generate_gamma_poisson_data(
N, M, K, graph_type, avg_degree, count_scale, a0, b0, c0, d0, seed=seed)
elif model_family == 'gaussian_gaussian':
sigma_U = config.get('sigma_U', 1.0)
sigma_V = config.get('sigma_V', 1.0)
sigma_x = config.get('sigma_x', 1.0)
edges, U_true, V_true, graph_edges = generate_gaussian_gaussian_data(
N, M, K, graph_type, avg_degree, sigma_U, sigma_V, sigma_x, seed=seed)
elif model_family == 'gaussian_gamma_map':
sigma_x = config.get('sigma_x', 1.0)
edges, U_true, V_true, graph_edges = generate_gaussian_gamma_data(
N, M, K, graph_type, avg_degree, a0, b0, c0, d0, sigma_x, seed=seed)
if len(edges) < 10:
print(f" WARNING: Only {len(edges)} edges generated, skipping config")
return []
graph_stats = compute_graph_stats([(e[0], e[1]) for e in edges], N, M)
print(f" Graph: {graph_stats['n_edges']} edges, mean user deg={graph_stats['user_degree_mean']:.1f}")
# Fit full model
print(f" Fitting full model...")
t_full_start = time.time()
if model_family == 'poisson_gamma':
model = PoissonGammaVI(N, M, K, a0, b0, c0, d0, max_iter=max_iter, tol=tol, seed=seed)
elif model_family == 'gaussian_gaussian':
model = GaussianGaussianVI(N, M, K, sigma_U=config.get('sigma_U', 1.0),
sigma_V=config.get('sigma_V', 1.0),
sigma_x=config.get('sigma_x', 1.0),
max_iter=max_iter, tol=tol, seed=seed)
elif model_family == 'gaussian_gamma_map':
model = GaussianGammaMAP(N, M, K, a0, b0, c0, d0,
sigma_x=config.get('sigma_x', 1.0),
lr=config.get('lr', 0.01),
max_iter=max_iter, tol=tol, seed=seed)
full_result = model.fit_full(edges)
full_params = full_result.params
t_full = time.time() - t_full_start
print(f" Full fit: {full_result.n_iterations} iters, {t_full:.1f}s, converged={full_result.converged}")
# Sample deletions
user_to_items, item_to_users, edge_dict = build_adjacency(edges, N, M)
deletion_samples = sample_deletions(edges, user_to_items, item_to_users, num_deletions, seed=seed)
print(f" Running {len(deletion_samples)} deletions...")
records = []
for del_idx, (edge_to_del, del_type) in enumerate(deletion_samples):
if del_idx % 5 == 0:
print(f" Deletion {del_idx+1}/{len(deletion_samples)} ({del_type})")
i_del, j_del, x_del = edge_to_del
# Exact deletion
t0 = time.time()
exact_result = model.fit_without_edge(edges, edge_to_del, init_params=full_params)
t_exact = time.time() - t0
exact_params = exact_result.params
# Local deletions for each radius
local_results = {}
local_params_by_radius = {}
for R in radii:
t0 = time.time()
local_result = model.fit_local(edges, edge_to_del, R, init_params=full_params)
local_results[R] = local_result
local_params_by_radius[R] = local_result.params
# Warm-start global
t0 = time.time()
ws_result = model.fit_warm_start_global(edges, edge_to_del, init_params=full_params)
# One-step downdate (only for poisson_gamma)
one_step_params = None
one_step_runtime = None
if model_family == 'poisson_gamma':
os_result = one_step_downdate_poisson_gamma(
edges, edge_to_del, full_params, N, M, K, a0, b0, c0, d0)
one_step_params = os_result.params
one_step_runtime = os_result.runtime_sec
# Compute all metrics
model_kwargs = {}
if model_family == 'poisson_gamma':
model_kwargs = {'a0': a0, 'b0': b0, 'c0': c0, 'd0': d0}
else:
model_kwargs = {'sigma_x': config.get('sigma_x', 1.0)}
metrics = compute_all_metrics(
full_params, exact_params, local_params_by_radius,
ws_result.params, one_step_params,
edge_to_del, edges, N, M, K,
model_family, model=model, radii=radii,
model_kwargs=model_kwargs)
# Build record
record = {
'run_id': run_id,
'config_id': config_id,
'dataset_type': 'synthetic',
'dataset_name': f'synthetic_{model_family}',
'model_family': model_family,
'inference_type': 'vi' if model_family != 'gaussian_gamma_map' else 'map',
'likelihood': 'poisson' if model_family == 'poisson_gamma' else 'gaussian',
'prior': 'gamma' if 'gamma' in model_family else 'gaussian',
'graph_type': graph_type,
'seed': seed,
'N': N,
'M': M,
'K': K,
'avg_degree': avg_degree,
'count_scale': count_scale if model_family == 'poisson_gamma' else None,
'prior_strength': prior_strength,
'deletion_edge': [int(i_del), int(j_del), float(x_del)],
'deletion_type': del_type,
'deletion_index': del_idx,
# Runtimes
'runtime_full': t_full,
'runtime_exact': t_exact,
'runtime_warm_start': ws_result.runtime_sec,
'runtime_one_step': one_step_runtime,
# Exact deletion convergence
'exact_converged': exact_result.converged,
'exact_iterations': exact_result.n_iterations,
'ws_converged': ws_result.converged,
'ws_iterations': ws_result.n_iterations,
}
# Add local runtimes
for R in radii:
record[f'runtime_local_R{R}'] = local_results[R].runtime_sec
record[f'local_R{R}_converged'] = local_results[R].converged
record[f'local_R{R}_iterations'] = local_results[R].n_iterations
# Add metrics
record.update(metrics)
# Flatten influence_by_distance for CSV compatibility
if 'influence_by_distance' in record:
for d_str, val in record['influence_by_distance'].items():
record[f'influence_d{d_str}'] = val
# Add regime label
record['regime'] = f"{graph_type}_{prior_strength}_deg{avg_degree}_cs{count_scale}"
# Extra config fields
if model_family == 'poisson_gamma':
record['a0'] = a0
record['b0'] = b0
record['c0'] = c0
record['d0'] = d0
if model_family in ('gaussian_gaussian', 'gaussian_gamma_map'):
record['sigma_x'] = config.get('sigma_x', 1.0)
if model_family == 'gaussian_gaussian':
record['sigma_U'] = config.get('sigma_U', 1.0)
record['sigma_V'] = config.get('sigma_V', 1.0)
records.append(record)
return records
def build_config_grid(grid_cfg, mode='pilot'):
"""Build list of configs from grid specification."""
cfg = grid_cfg[mode] if mode in grid_cfg else grid_cfg.get('pilot', grid_cfg)
configs = []
N = cfg['N']
M = cfg['M']
radii = cfg.get('radii', [1, 2, 3, 4])
num_del = cfg.get('num_deletions_per_config', 20)
seed = cfg.get('seed', 42)
for K in cfg['K_values']:
for gt in cfg['graph_types']:
for deg_name, deg_val in cfg['avg_degree_levels'].items():
for cs_name, cs_val in cfg['count_scale_levels'].items():
for ps_name, ps_cfg in cfg['prior_strength_configs'].items():
configs.append({
'N': N, 'M': M, 'K': K,
'graph_type': gt,
'avg_degree': deg_val,
'avg_degree_label': deg_name,
'count_scale': cs_val,
'count_scale_label': cs_name,
'prior_strength': ps_name,
'prior_config': ps_cfg,
'num_deletions': num_del,
'radii': radii,
'seed': seed,
'model_family': 'poisson_gamma',
})
return configs
def build_model_family_grid(grid_cfg):
"""Build config grid for model-family ablation."""
cfg = grid_cfg.get('model_family_ablation', {})
if not cfg:
return []
configs = []
N = cfg.get('N', 200)
M = cfg.get('M', 200)
radii = cfg.get('radii', [1, 2, 3, 4])
num_del = cfg.get('num_deletions_per_config', 30)
seed = cfg.get('seed', 42)
graph_types = cfg.get('graph_types', ['bounded_degree', 'erdos_renyi', 'power_law'])
deg_values = cfg.get('avg_degree_values', [5, 15])
K_values = cfg.get('K_values', [5, 10])
# Poisson-Gamma configs
pg_cfg = cfg.get('poisson_gamma', {})
for K in K_values:
for gt in graph_types:
for deg in deg_values:
for cs_name, cs_val in pg_cfg.get('count_scale_levels', {'medium': 1.0}).items():
for ps_name, ps_dict in pg_cfg.get('prior_configs', {'strong': {'a0': 1.0, 'b0': 1.0, 'c0': 1.0, 'd0': 1.0}}).items():
configs.append({
'N': N, 'M': M, 'K': K,
'graph_type': gt, 'avg_degree': deg,
'count_scale': cs_val, 'count_scale_label': cs_name,
'prior_strength': ps_name, 'prior_config': ps_dict,
'num_deletions': num_del, 'radii': radii, 'seed': seed,
'model_family': 'poisson_gamma',
})
# Gaussian-Gaussian configs
gg_cfg = cfg.get('gaussian_gaussian', {})
for K in K_values:
for gt in graph_types:
for deg in deg_values:
for sx_name, sx_val in gg_cfg.get('sigma_x_values', {'medium_noise': 1.0}).items():
for sp_name, sp_val in gg_cfg.get('sigma_prior_values', {'strong_prior': 0.5}).items():
configs.append({
'N': N, 'M': M, 'K': K,
'graph_type': gt, 'avg_degree': deg,
'sigma_x': sx_val, 'sigma_x_label': sx_name,
'sigma_U': sp_val, 'sigma_V': sp_val,
'prior_strength': sp_name,
'num_deletions': num_del, 'radii': radii, 'seed': seed,
'model_family': 'gaussian_gaussian',
})
# Gaussian-Gamma MAP configs
ggm_cfg = cfg.get('gaussian_gamma_map', {})
for K in K_values:
for gt in graph_types:
for deg in deg_values:
for sx_name, sx_val in ggm_cfg.get('sigma_x_values', {'medium_noise': 1.0}).items():
for gp_name, gp_dict in ggm_cfg.get('gamma_prior_strength', {'strong': {'a0': 2.0, 'b0': 2.0, 'c0': 2.0, 'd0': 2.0}}).items():
configs.append({
'N': N, 'M': M, 'K': K,
'graph_type': gt, 'avg_degree': deg,
'sigma_x': sx_val, 'sigma_x_label': sx_name,
'prior_strength': gp_name, 'prior_config': gp_dict,
'num_deletions': num_del, 'radii': radii, 'seed': seed,
'model_family': 'gaussian_gamma_map',
'lr': 0.005,
})
return configs
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--config', type=str, default='config/synthetic_grid.yaml')
parser.add_argument('--mode', type=str, default='pilot', choices=['pilot', 'full', 'model_family'])
parser.add_argument('--max_configs', type=int, default=None, help='Limit number of configs for testing')
args = parser.parse_args()
with open(args.config) as f:
grid_cfg = yaml.safe_load(f)
if args.mode == 'model_family':
configs = build_model_family_grid(grid_cfg)
else:
configs = build_config_grid(grid_cfg, args.mode)
if args.max_configs:
configs = configs[:args.max_configs]
print(f"Running {len(configs)} configurations in {args.mode} mode")
output_dir = ensure_dir('results/raw')
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
output_file = os.path.join(output_dir, f'synthetic_{args.mode}_{timestamp}.jsonl')
all_records = []
for cfg_idx, config in enumerate(configs):
print(f"\n>>> Config {cfg_idx+1}/{len(configs)}")
try:
records = run_single_config(config, mode=args.mode)
all_records.extend(records)
# Save incrementally
save_jsonl(records, output_file)
print(f" Saved {len(records)} records (total: {len(all_records)})")
except Exception as e:
print(f" ERROR in config {cfg_idx+1}: {e}")
import traceback
traceback.print_exc()
# Log to debug
debug_entry = {
'config_index': cfg_idx,
'config': {k: str(v) for k, v in config.items()},
'error': str(e),
'timestamp': datetime.now().isoformat(),
}
debug_path = 'debug_errors.jsonl'
with open(debug_path, 'a') as f:
f.write(json.dumps(debug_entry) + '\n')
print(f"\n{'='*60}")
print(f"Completed. Total records: {len(all_records)}")
print(f"Output: {output_file}")
return output_file
if __name__ == '__main__':
main()
|