environment_state_actor.py

import ray
from datetime import datetime

class EnvironmentState:
    def __init__(self, episode, step):
        self.timestamp = datetime.utcnow()
        self.episode = episode
        self.step = step
        self.reward = 0
        self.llm_preview = ''
        self.llm_responses = []
        self.tts_raw_chunk_ids = [] 

    def __str__(self):
        state = ', '.join(f'{k}={v}' for k, v in self.__dict__.items() if k not in {'episode', 'step', 'timestamp', 'reward'})
        return f'episode={self.episode}, step={self.step}, timestamp={self.timestamp}, \nreward={self.reward}\nstate=({state})'


@ray.remote
class EnvironmentStateActor:
    def __init__(self):
        self.episode = 0
        self.step = 0
        self.state = None
        self.reset_episode()

    def reset_episode(self):
        self.episode += 1
        self.step = 0
        self.state = EnvironmentState(self.episode, self.step)
        return self.state

    def begin_next_step(self)->EnvironmentState:
        previous_state = self.state
        self.step += 1
        self.state = EnvironmentState(self.episode, self.step)
        return previous_state

    def add_reward(self, reward):
        self.state.reward += reward

    def set_llm_preview(self, llm_preview):
        self.state.llm_preview = llm_preview

    def add_llm_response_and_clear_llm_preview(self, llm_response):
        self.state.llm_responses.append(llm_response)
        self.state.llm_preview = ''

    def add_tts_raw_chunk_id(self, chunk_id):
        self.state.tts_raw_chunk_ids.append(chunk_id)

    def get_state(self)->EnvironmentState:
        return self.state