Merge branch 'master' of github.com:ccnmaastricht/dexterous-robot-hand

README.md

@@ -64,32 +64,14 @@ Where ID is the agent's ID given when its created (`train.py` prints this outt,
 To train agents with custom models, environments, etc. you write your own script. The following is a minimal example:
 
 ```python
-from angorapy.agent.ppo_agent import PPOAgent
-from angorapy.common.policies import BetaPolicyDistribution
-from angorapy.common.transformers import RewardNormalizationTransformer, StateNormalizationTransformer
 from angorapy.common.wrappers import make_env
 from angorapy.models import get_model_builder
+from angorapy.agent.ppo_agent import PPOAgent
 
-wrappers = [StateNormalizationTransformer, RewardNormalizationTransformer]
-env = make_env("LunarLanderContinuous-v2", reward_config=None, transformers=wrappers)
-
-# make policy distribution
-distribution = BetaPolicyDistribution(env)
-
-# the agent needs to create the model itself, so we build a method that builds a model
-build_models = get_model_builder(model="simple", model_type="ffn", shared=False)
-
-# given the model builder and the environment we can create an agent
-agent = PPOAgent(build_models, env, horizon=1024, workers=12, distribution=distribution)
-
-# let's check the agents ID, so we can find its saved states after training
-print(f"My Agent's ID: {agent.agent_id}")
-
-# ... and then train that agent for n cycles
-agent.drill(n=100, epochs=3, batch_size=64)
-
-# after training, we can save the agent for analysis or the like
-agent.save_agent_state()
+env = make_env("LunarLanderContinuous-v2")
+model_builder = get_model_builder("simple", "ffn")
+agent = PPOAgent(model_builder, env)
+agent.drill(100, 10, 512)
 ```
 
 For more details, consult the [examples](examples).