| """V7 counterfactual-trained structured spectral transport benchmark.""" |
| from __future__ import annotations |
| import argparse,csv,json |
| from pathlib import Path |
| from statistics import mean,stdev |
| import torch, numpy as np |
| from spectral_world_models.hard_dynamics import HardDynamicsConfig,generate_hard_benchmark_npz |
| from spectral_world_models.train import train_one_model |
| from spectral_world_models.counterfactual import evaluate_counterfactual |
| from spectral_world_models.models import build_model |
| DEFAULT_MODELS=['neural_operator','swm_selective','swm_structured','swm_structured_cf','no_spectral_transition'] |
| def stats(x): return mean(x),stdev(x) if len(x)>1 else 0. |
| def main(): |
| ap=argparse.ArgumentParser(); ap.add_argument('--epochs',type=int,default=3); ap.add_argument('--batch-size',type=int,default=64); ap.add_argument('--seeds',type=int,nargs='+',default=[0,1,2,3,4]); ap.add_argument('--horizons',type=int,nargs='+',default=[5,10,20,30,50,100]); ap.add_argument('--cf-horizon',type=int,default=30); ap.add_argument('--device',default='auto'); ap.add_argument('--cf-train-horizon',type=int,default=8); ap.add_argument('--cf-pairs',type=int,default=64); ap.add_argument('--cf-every',type=int,default=4); ap.add_argument('--lambda-cf-dir',type=float,default=0.02); ap.add_argument('--lambda-cf-mag',type=float,default=0.005); ap.add_argument('--lambda-cf-branch',type=float,default=0.10); ap.add_argument('--models',nargs='+',default=DEFAULT_MODELS); a=ap.parse_args() |
| root=Path(__file__).resolve().parents[1]; data=root/'data/hard_controlled_dynamics_v6.npz'; rollout=root/f'data/hard_controlled_dynamics_v6_rollout_h{max(a.horizons)+1}.npz'; out=root/'results/v7_counterfactual_training'; out.mkdir(parents=True,exist_ok=True) |
| if not data.exists(): generate_hard_benchmark_npz(data,HardDynamicsConfig()) |
| if not rollout.exists(): generate_hard_benchmark_npz(rollout,HardDynamicsConfig(seq_len=max(a.horizons)+1,train_sequences=1,val_sequences=1,test_sequences=96,seed=7017)) |
| device=('cuda' if torch.cuda.is_available() else 'cpu') if a.device=='auto' else a.device; grouped={m:[] for m in a.models}; raw=[] |
| for seed in a.seeds: |
| for m in a.models: |
| print(f'\n=== seed={seed} model={m} ==='); sd=out/f'seed_{seed}'; r=train_one_model(m,data,sd,epochs=a.epochs,batch_size=a.batch_size,seed=seed,device=device,rollout_data_path=rollout,rollout_horizons=tuple(a.horizons),cf_train_horizon=a.cf_train_horizon,cf_pairs=a.cf_pairs,cf_every=a.cf_every,lambda_cf_dir=a.lambda_cf_dir,lambda_cf_mag=a.lambda_cf_mag,lambda_cf_branch=a.lambda_cf_branch) |
| ck=torch.load(sd/f'{m}.pt',map_location=device); model=build_model(m).to(device); model.load_state_dict(ck['model']); cf=evaluate_counterfactual(model,torch.device(device),horizon=a.cf_horizon,seed=606+seed); r['counterfactual']=cf; grouped[m].append(r) |
| row={'seed':seed,'model':m,'params':r['params'],'test_mse':r['test']['mse'],'test_text_acc':r['test']['text_acc'],**cf} |
| for h in a.horizons: |
| rr=r['rollout_by_horizon'][str(h)]; row[f'rollout_mse_h{h}']=rr['rollout_mse']; row[f'rollout_text_acc_h{h}']=rr['rollout_text_acc']; row[f'rollout_latent_cosine_h{h}']=rr['rollout_latent_cosine'] |
| row.update(r.get('stability_diagnostics',{})); raw.append(row) |
| fields=[] |
| for r in raw: |
| for k in r: |
| if k not in fields: fields.append(k) |
| with (out/'metrics_by_seed.csv').open('w',newline='') as f: w=csv.DictWriter(f,fieldnames=fields);w.writeheader();w.writerows(raw) |
| summary=[] |
| for m,runs in grouped.items(): |
| row={'model':m,'params':runs[0]['params'],'n_seeds':len(runs)} |
| getters={'test_mse':lambda r:r['test']['mse'],'test_text_acc':lambda r:r['test']['text_acc'],'cf_final_image_mse':lambda r:r['counterfactual']['cf_final_image_mse'],'cf_effect_vector_cosine':lambda r:r['counterfactual']['cf_effect_vector_cosine'],'cf_effect_magnitude_ratio':lambda r:r['counterfactual']['cf_effect_magnitude_ratio']} |
| for h in a.horizons: |
| for q in ['rollout_mse','rollout_text_acc','rollout_latent_cosine']: getters[f'{q}_h{h}']=lambda r,q=q,h=h:r['rollout_by_horizon'][str(h)][q] |
| for k,g in getters.items(): mu,sd=stats([float(g(r)) for r in runs]);row[k+'_mean']=mu;row[k+'_std']=sd |
| dkeys=set().union(*(r.get('stability_diagnostics',{}).keys() for r in runs)) |
| for k in dkeys: |
| vals=[float(r['stability_diagnostics'][k]) for r in runs if k in r.get('stability_diagnostics',{})];mu,sd=stats(vals);row[k+'_mean']=mu;row[k+'_std']=sd |
| summary.append(row) |
| fields=[] |
| for r in summary: |
| for k in r: |
| if k not in fields: fields.append(k) |
| with (out/'metrics_summary.csv').open('w',newline='') as f:w=csv.DictWriter(f,fieldnames=fields);w.writeheader();w.writerows(summary) |
| with (out/'run_config.json').open('w') as f:json.dump(vars(a)|{'device_resolved':device},f,indent=2) |
| print('\nWrote',out/'metrics_summary.csv') |
| if __name__=='__main__':main() |
|
|
