Package the project

README.md

@@ -57,12 +57,12 @@ Prediction is sampled after 110K iterations, and I only implemented one-step tra
 * MacOS 10.12 or Ubuntu 16.04
 * PyTorch v0.4.0
 * Python 3.6, 3.5 or 2.7 (deprecated)
-* Core dependencies: `pip install -r requirements.txt`
+* Core dependencies: `pip install -e .`
 * Optional: [Roboschool](https://github.com/openai/roboschool), [DeepMind Control Suite](https://github.com/deepmind/dm_control)+[DMControl2Gym](dm_control2gym)
 
 # Usage
 
-```main.py``` contains examples for all the implemented algorithms
+```examples.py``` contains examples for all the implemented algorithms
 
 # References
 * [Human Level Control through Deep Reinforcement Learning](https://www.nature.com/nature/journal/v518/n7540/full/nature14236.html)

deep_rl/__init__.py

@@ -0,0 +1,5 @@
+from .agent import *
+from .component import *
+from .model import *
+from .network import *
+from .utils import *

deep_rl/agent/A2C_agent.py

@@ -4,14 +4,9 @@
 # declaration at the top                                              #
 #######################################################################
 
-import numpy as np
-from network import *
-from utils import *
-from component import *
+from ..network import *
+from ..component import *
 from .BaseAgent import *
-import pickle
-import os
-import time
 
 class A2CAgent(BaseAgent):
     def __init__(self, config):

deep_rl/agent/CategoricalDQN_agent.py

@@ -4,14 +4,10 @@
 # declaration at the top                                              #
 #######################################################################
 
-from network import *
-from component import *
-from utils import *
-import numpy as np
+from ..network import *
+from ..component import *
+from ..utils import *
 import time
-import os
-import pickle
-import torch
 from .BaseAgent import *
 
 class CategoricalDQNAgent(BaseAgent):

deep_rl/agent/DDPG_agent.py

@@ -4,14 +4,8 @@
 # declaration at the top                                              #
 #######################################################################
 
-import numpy as np
-import torch.multiprocessing as mp
-from network import *
-from utils import *
-from component import *
-import pickle
-import os
-import time
+from ..network import *
+from ..component import *
 from .BaseAgent import *
 
 class DDPGAgent(BaseAgent):

deep_rl/agent/DQN_agent.py

@@ -4,14 +4,10 @@
 # declaration at the top                                              #
 #######################################################################
 
-from network import *
-from component import *
-from utils import *
-import numpy as np
+from ..network import *
+from ..component import *
+from ..utils import *
 import time
-import os
-import pickle
-import torch
 from .BaseAgent import *
 
 class DQNAgent(BaseAgent):

deep_rl/agent/NStepDQN_agent.py

@@ -4,14 +4,9 @@
 # declaration at the top                                              #
 #######################################################################
 
-from network import *
-from component import *
-from utils import *
-import numpy as np
-import time
-import os
-import pickle
-import torch
+from ..network import *
+from ..component import *
+from ..utils import *
 from .BaseAgent import *
 
 class NStepDQNAgent(BaseAgent):

deep_rl/agent/PPO_agent.py

@@ -4,14 +4,8 @@
 # declaration at the top                                              #
 #######################################################################
 
-import numpy as np
-import torch.multiprocessing as mp
-from network import *
-from utils import *
-from component import *
-import pickle
-import os
-import time
+from ..network import *
+from ..component import *
 from .BaseAgent import *
 
 class PPOAgent(BaseAgent):

deep_rl/agent/QuantileRegressionDQN_agent.py

@@ -4,14 +4,10 @@
 # declaration at the top                                              #
 #######################################################################
 
-from network import *
-from component import *
-from utils import *
-import numpy as np
+from ..network import *
+from ..component import *
+from ..utils import *
 import time
-import os
-import pickle
-import torch
 from .BaseAgent import *
 
 class QuantileRegressionDQNAgent(BaseAgent):

deep_rl/component/atari_wrapper.py

@@ -1,7 +1,6 @@
 # based on https://github.com/openai/baselines/blob/master/baselines/common/atari_wrappers.py
 
 import numpy as np
-from collections import deque
 import gym
 from gym import spaces
 from gym.spaces import Box

deep_rl/component/replay.py

@@ -5,9 +5,6 @@
 #######################################################################
 
 import numpy as np
-import torch
-import random
-import torch.multiprocessing as mp
 
 class Replay:
     def __init__(self, memory_size, batch_size):

deep_rl/component/task.py

@@ -3,15 +3,11 @@
 # Permission given to modify the code as long as you keep this        #
 # declaration at the top                                              #
 #######################################################################
-import gym
-import sys
-import numpy as np
 from .atari_wrapper import *
 import multiprocessing as mp
 import sys
 from .bench import Monitor
-from utils import *
-import datetime
+from ..utils import *
 import uuid
 
 class BaseTask:

deep_rl/model/action_conditional_video_prediction.py

@@ -6,20 +6,14 @@
 
 __all__ = ['acvp_train']
 
-import torch
-from torch.autograd import Variable
-import torch.nn as nn
-import torch.nn.functional as F
-import numpy as np
-import pickle
 import torchvision
 from skimage import io
 from collections import deque
 import gym
 import torch.optim
-from utils import *
+from deep_rl.utils import *
 from tqdm import tqdm
-from network import *
+from deep_rl.network import *
 
 class Network(nn.Module, BaseNet):
     def __init__(self, num_actions, gpu=0):

deep_rl/model/dataset.py

@@ -6,10 +6,9 @@
 
 __all__ = ['generate_dataset']
 
-import logging
-from agent import *
-from component import *
-from utils import *
+from ..agent import *
+from ..component import *
+from deep_rl.utils import *
 from skimage import io
 
 def episode(agent, task):

deep_rl/network/network_utils.py

@@ -7,7 +7,6 @@
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-import numpy as np
 
 class BaseNet:
     def set_gpu(self, gpu):

deep_rl/utils/misc.py

@@ -8,7 +8,6 @@ import numpy as np
 import pickle
 import os
 import datetime
-import uuid
 import torch
 try:
     # python >= 3.5

deep_rl/utils/normalizer.py

@@ -3,7 +3,6 @@
 # Permission given to modify the code as long as you keep this        #
 # declaration at the top                                              #
 #######################################################################
-import torch
 import numpy as np
 
 class BaseNormalizer:

deep_rl/utils/plot.py

@@ -1,7 +1,7 @@
 # Adapted from  https://github.com/openai/baselines/blob/master/baselines/results_plotter.py
 
 import numpy as np
-import component
+from ..component import *
 import os
 import re
 

examples.py

@@ -4,11 +4,7 @@
 # declaration at the top                                              #
 #######################################################################
 
-import logging
-from agent import *
-from component import *
-from utils import *
-from model import *
+from deep_rl import *
 
 ## cart pole
 

requirements.txt

@@ -1,12 +0,0 @@
-torch>=0.4.0
-torchvision>=0.2.1
-gym>=0.10.5
-atari-py>=0.1.1
-opencv-python>=3.4.0.12
-tensorboardX==1.1
-scikit-image>=0.13.1
-tqdm>=4.23.0
-pandas>=0.22.0
-pathlib2>=1.0.1;python_version <= '2.7'
-pathlib>=1.0.1;python_version >= '3.5'
-seaborn>=0.8.1

setup.py

@@ -0,0 +1,31 @@
+from setuptools import setup, find_packages
+import sys
+
+if sys.version.startswith('2.7'):
+    pathlib = 'pathlib2>=1.0.1'
+elif sys.version.startswith('3.5') or sys.version.startswith('3.6'):
+    pathlib = 'pathlib>=1.0.1'
+else:
+    raise Exception('Only Python 2.7, 3.5 and 3.6 are supported')
+
+setup(name='deep_rl',
+      packages=[package for package in find_packages()
+                if package.startswith('deep_rl')],
+      install_requires=[
+          'torch>=0.4.0',
+          'torchvision>=0.2.1',
+          'gym>=0.10.5',
+          'atari-py>=0.1.1',
+          'opencv-python>=3.4.0.12',
+          'tensorboardX==1.1',
+          'scikit-image>=0.13.1',
+          'tqdm>=4.23.0',
+          'pandas>=0.22.0',
+          'seaborn>=0.8.1',
+          pathlib
+      ],
+      description="Highly modularized implementation of popular deep RL algorithms",
+      author="Shangtong Zhang",
+      url='https://github.com/ShangtongZhang/DeepRL',
+      author_email="zhangshangtong.cpp@gmail.com",
+      version="0.2")