To train / test a DQN without experience replay, use script_v1.py

with experience replay, use script_v2.py

To train / test on different envs, modify the string pass to the initializer of EnvWrapper

To use a linear network, set 'v1' in the initializer of DeepQN,

       triple-layer DQN, set 'v3'
       Conv-DQN,       set  'v5'

An easy version:

1. Want experience replay? Yes --> use script_v1.py No --> use script_v2.py

2. Want which env to be played? CartPole --> env = EnvWrapper('CartPole-v0') --> continue to step 3 Car --> env = EnvWrapper('MountainCar-v0') --> continue to step 3 SpaceInv --> env = EnvWrapper('SpaceInvaders-v0') --> continue to step 4

3. Want linear or non-linear DQN? Linear --> qn = DeepQN(..., type='v1', ...) NonLinear --> qn = DeepQN(..., type='v3', ...) Dual NonL --> qn = DeepQN(..., type='v5', ...)

4. Use Conv DQN --> qn = DeepQN(..., type='v4', ...)