"""STER reproduction driver — runs the official 3dSAGER matching pipeline.
Loads pre-computed property dicts if present (skips the ~250s property step).
Overrides config programmatically (no source edits)."""
import sys, warnings, time, json
warnings.filterwarnings("ignore")

import config
config.Models.model_list = ['BaggingClassifier', 'RandomForestClassifier']
config.Constants.dataset_name = 'Hague'
config.Constants.evaluation_mode = 'matching'
config.Constants.dataset_size_version = 'small'
config.Constants.seeds_num = 1
config.Constants.load_object_dict = True
config.Constants.load_property_dict = True     # <-- reuse saved property dicts (fast)
config.Constants.save_property_dict = True
config.Constants.load_dataset_dict = False
config.Constants.save_dataset_dict = False
config.Constants.save_object_dict = False

from utils import define_logger, print_config, generate_final_result_csv
from pipelines import PipelineManager

class Args: pass
args = Args()
args.dataset_name = 'Hague'; args.evaluation_mode = 'matching'
args.run_preparatory_phase = config.TrainingPhase.run_preparatory_phase
args.blocking_method = config.Blocking.blocking_method
args.seeds_num = 1; args.dataset_size_version = 'small'
args.vector_normalization = True; args.sdr_factor = False; args.neg_samples_num = 2
args.bkafi_criterion = 'feature_importance'; args.run_blocker_train = False
args.matching_cands_generation = 'blocking-based'; args.contamination_mode = False

logger = define_logger()
print_config(logger, args)
t0 = time.time()
result_dict = {}
for seed in range(1, args.seeds_num + 1):
    logger.info(f"Seed: {seed}")
    pm = PipelineManager(seed, logger, args)
    result_dict[seed] = pm.result_dict
    print("\n==== RESULT DICT (seed", seed, ") ====", flush=True)
    def _san(o):
        try: json.dumps(o); return o
        except Exception: return str(o)
    print(json.dumps(result_dict[seed], indent=2, default=_san)[:4000], flush=True)
print(f"\nTotal time: {time.time()-t0:.1f}s", flush=True)