Sheep training flock of 10 fix?

2026-04-24 15:14:45 +01:00
parent bdbe8ba1de
commit 44b2788e78
1 changed files with 4 additions and 6 deletions
@@ -98,11 +98,8 @@ def train_stage(n_sheep, steps, n_envs=4, prev_model=None, prev_vecnorm=None):
        vn = VecNormalize(train_env, norm_obs=True, norm_reward=True, clip_obs=10.0)

    if prev_model is not None:
-        model = PPO.load(prev_model, env=vn,
-                         learning_rate=3e-4, n_steps=2048, batch_size=256,
-                         n_epochs=10, gamma=0.995, gae_lambda=0.95,
-                         clip_range=0.2, ent_coef=0.005, vf_coef=0.5,
-                         max_grad_norm=0.5)
+        model = prev_model
+        model.set_env(vn)
    else:
        model = PPO(
            "MlpPolicy", vn,
@@ -113,7 +110,8 @@ def train_stage(n_sheep, steps, n_envs=4, prev_model=None, prev_vecnorm=None):
            verbose=1,
        )

-    model.learn(total_timesteps=steps, reset_num_timesteps=(prev_model is None))
+    model.learn(total_timesteps=steps, reset_num_timesteps=(prev_model is None),
+                tb_log_name="ppo_smoke")
    return model, vn