A2C from stable_baselines with torcs front end

Author: siddhya

test.py

@@ -0,0 +1,30 @@
+import torch as th
+import gym_torcs
+import gym
+import sys
+import os
+
+from stable_baselines3 import A2C
+
+#env = gym.make('CartPole-v1')
+env = gym_torcs.TorcsEnv()
+policy_kwargs = dict(activation_fn=th.nn.ReLU,
+                     net_arch=[dict(pi=[128, 128], vf=[128, 128])])
+
+if os.path.exists('a2c_torcs.zip'):
+  print('Using saved model')
+  model = A2C.load('a2c_torcs', env=env)
+else:
+  model = A2C('MlpPolicy', env, policy_kwargs=policy_kwargs, verbose=1)
+model.learn(total_timesteps=1000, log_interval=5)
+print('Learning done')
+model.save('a2c_torcs')
+env.end()
+sys.exit()
+obs = env.reset()
+for i in range(1000):
+    action, _state = model.predict(obs, deterministic=True)
+    obs, reward, done, info = env.step(action)
+    env.render()
+    if done:
+      obs = env.reset()