rl_algo_multi.py

def multi_ddpg(n_episodes=5000, max_t=2000):
    init_time = time.time();
    scores_deque = deque(maxlen=100);
    scores = []
    max_score = -np.Inf;
    
    for i_episode in range(1, n_episodes+1):
        ep_init_time = time.time();
        env_info = env.reset(train_mode=True)[brain_name];
        states = env_info.vector_observations;
        agent.reset();
        agent_scores = np.zeros(num_agents);
        
        for t_step in range(max_t):
            actions = agent.act(states);
            next_env_info = env.step(actions)[brain_name];   
            next_states, rewards, dones = next_env_info.vector_observations, next_env_info.rewards, next_env_info.local_done;
            agent.step(states, actions, rewards, next_states, dones);           
            agent_scores = agent_scores + rewards;        
            states = next_states;    
            if np.any(dones):
                break;
                
        best_score = np.max(agent_scores);
        scores_deque.append(best_score);
        scores.append(best_score);
        epi_score = np.mean(scores_deque);
        sys.stdout.write('\rEpisode %d\t Done in %.2f/%.2f\t Scores: %s\tAvg Score %.2f'%(i_episode, (time.time() - ep_init_time), (time.time() - init_time), ['%.2f'%sc for sc in agent_scores], np.mean(scores_deque)))
        
        if i_episode > 100 and epi_score >=0.5:
            if epi_score > max_score:
                max_score = epi_score;            
                agent.save(i_episode);
            sys.stdout.write('\nEnvironment Solved: %d\t Done in %.2f/%.2f\t Scores: %s\tAvg Score %.2f'%(i_episode, (time.time() - ep_init_time), (time.time() - init_time), ['%.2f'%sc for sc in agent_scores], np.mean(scores_deque)))
    return scores;