holy fuking shit

2026-09-12 01:56:34 +08:00
parent 36f354736c
commit d975309a81
83 changed files with 2748 additions and 896 deletions
@@ -9,7 +9,7 @@ np = pytest.importorskip('numpy')
 pytest.importorskip('gymnasium')

 from mcm26b.rl.env import LocalizationEnv
-from mcm26b.rl.features import action_mask, build_observation, candidate_actions
+from mcm26b.rl.features import build_observation, candidate_actions


 def test_training_observation_matches_runtime_channel_belief():
@@ -85,10 +85,12 @@ def test_bc_and_ppo_forward_mapping_agree():
    spec.loader.exec_module(module)
    torch.set_num_threads(1)
    env = LocalizationEnv()
-    ppo = PPO('MlpPolicy',env,policy_kwargs={'net_arch':[128,128]},device='cpu',n_steps=16,batch_size=8)
+    ppo = PPO('MlpPolicy',env,policy_kwargs={'net_arch':[128,128]},device='cpu',
+              n_steps=16,batch_size=8)
    bc = module.BCMlp()
    state = bc.net.state_dict()
-    ppo.policy.mlp_extractor.policy_net.load_state_dict({k:v for k,v in state.items() if not k.startswith('4.')})
+    ppo.policy.mlp_extractor.policy_net.load_state_dict(
+        {k:v for k,v in state.items() if not k.startswith('4.')})
    ppo.policy.action_net.load_state_dict({'weight':state['4.weight'],'bias':state['4.bias']})
    x = torch.randn(10,401)
    torch.testing.assert_close(bc(x),ppo.policy.action_net(ppo.policy.mlp_extractor.policy_net(x)))