evaluate.py now expects a environment config file, RL algorithms are saved based on average return and also saving last model
