holy fuking shit

scripts/summarize_rl_revision.py

@@ -12,7 +12,8 @@ from mcm26b.rl.features import build_observation
 
 def main():
     root = Path('output/rl/review-20260911')
-    records = json.loads((root/'evaluation/validation/final/results.json').read_text(encoding='utf-8'))
+    records = json.loads((root/'evaluation/validation/final/results.json')
+                         .read_text(encoding='utf-8'))
     rng = np.random.default_rng(20260911)
     comparisons = {}
     for problem in (3,4):
@@ -46,13 +47,16 @@ def main():
     bc = {}
     for problem in (3,4):
         # 使用旧检查点和原始专家输入，比对原ReLU网络与拷贝后的Tanh网络。
-        checkpoint = torch.load(f'output/rl/bc/q{problem}_bc.pt',map_location='cpu',weights_only=True)
+        checkpoint = torch.load(f'output/rl/bc/q{problem}_bc.pt',map_location='cpu',
+                               weights_only=True)
         data = np.load(f'output/rl/bc/q{problem}_expert.npz',allow_pickle=False)
         x = torch.from_numpy(data['obs'][::max(1,len(data['obs'])//1000)].copy())
-        def network(activation):
+
+        def network(activation, checkpoint=checkpoint):
             net = torch.nn.Sequential(torch.nn.Linear(401,128),activation(),
                                       torch.nn.Linear(128,128),activation(),torch.nn.Linear(128,60))
-            net.load_state_dict({k.removeprefix('net.'):v for k,v in checkpoint['state_dict'].items()})
+            state = {k.removeprefix('net.'):v for k,v in checkpoint['state_dict'].items()}
+            net.load_state_dict(state)
             return net
         with torch.no_grad():
             before,after=network(torch.nn.ReLU)(x),network(torch.nn.Tanh)(x)
@@ -63,9 +67,11 @@ def main():
                   limitations='置信区间仅重采样固定模型的场景，不覆盖训练随机种子不确定性。',
                   code_sha256={str(p):hashlib.sha256(p.read_bytes()).hexdigest()
                                for p in sorted(Path('src/mcm26b/rl').glob('*.py'))})
-    (root/'review_evidence.json').write_text(json.dumps(report,ensure_ascii=False,indent=2),encoding='utf-8')
+    (root/'review_evidence.json').write_text(
+        json.dumps(report,ensure_ascii=False,indent=2),encoding='utf-8')
     print(json.dumps(dict(comparisons=comparisons,feature_probe=features,bc_probe=bc,
-                         official=[dict(case=r['case_code'],passed=r['passed']) for r in official]),indent=2))
+                         official=[dict(case=r['case_code'],passed=r['passed'])
+                                   for r in official]),indent=2))
 
 
 if __name__=='__main__':