How to evaluate the trained agent can save money than that of basic E+ model

[agl71]

Dear @mechyai

Thank you for your great work!
After running .\Current_Prototype\experiment_runner.py, I found that the value of RTP HVAC Cost Metric(2.106) after training process in the _exp_results_log.txt is bigger than that of Baseline value(1.054) .

1. How to understand the result?I think that when the model works well,the value of RTP HVAC Cost Metric after training process should be lower than that of baseline value. Does the result shows that the model don't work well?
2. How to evaluate the effect of a trained agent which can save money than that of no trained E+ model(baseline result?)
3. I found that the trained dqn model can not be passed into the agent when 'exploit_only=True' set in the experiment_params_dict in .\Current_Prototype\experiment_runner.py. Therefore, the trained dqn model can not be used in this case, is it ture?
4. In experiment_runner.py,the train_period is set into(4/21 - 4/30), and the test_period is set into(5/1 -
   5/31), could you tell me why train_period is shorter than test_period ? why not in opposite?
5. How to evaulate that a model has been trained well by metric some values such as reward cumulative ,loss and epsilon?

Best regards;