File structure:
----cdhrl_train.py: main function
----DQN.py: DQN algorithm
----SCM.py: Structural Causal Model defination and training functions
----HRL.py: Goal-based hierarchical policy defination and training functions
----models.py: The network defination of agent
----mc.py: 2D-Minecraft environment wrapper
----utils.py utils functions
----minecraft: implementation of 2D-Minecraft
----pretrain_models: pretrained goal-based hierarchical policy models of 2D-Minecraft

Commands(use mpi to run parallel):
train task policy of 2D-Minecraft:
    mpiexec -n 4 python -u cdhrl_train.py --model_path YourModelAndLogsPath --train_task True
pretrain goal-based hierarchical policy:
    mpiexec -n 16 python -u cdhrl_train.py --model_path YourModelAndLogsPath

