lstm network builders using tf lstm

.benchmark_pattern

@@ -1 +1 @@
-
+ppo2

.gitignore

@@ -5,7 +5,6 @@
 .pytest_cache
 .DS_Store
 .idea
-.coverage
 
 # Setuptools distribution and build folders.
 /dist/

README.md

@@ -139,4 +139,3 @@ To cite this repository in publications:
       journal = {GitHub repository},
       howpublished = {\url{https://github.com/openai/baselines}},
     }
-

baselines/common/atari_wrappers.py

@@ -156,7 +156,7 @@ class FrameStack(gym.Wrapper):
         self.k = k
         self.frames = deque([], maxlen=k)
         shp = env.observation_space.shape
-        self.observation_space = spaces.Box(low=0, high=255, shape=(shp[0], shp[1], shp[2] * k), dtype=env.observation_space.dtype)
+        self.observation_space = spaces.Box(low=0, high=255, shape=(shp[0], shp[1], shp[2] * k), dtype=np.uint8)
 
     def reset(self):
         ob = self.env.reset()
@@ -176,7 +176,6 @@ class FrameStack(gym.Wrapper):
 class ScaledFloatFrame(gym.ObservationWrapper):
     def __init__(self, env):
         gym.ObservationWrapper.__init__(self, env)
-        self.observation_space = gym.spaces.Box(low=0, high=1, shape=env.observation_space.shape, dtype=np.float32)
 
     def observation(self, observation):
         # careful! This undoes the memory optimization, use

baselines/common/models.py

@@ -138,7 +138,7 @@ def conv_only(convs=[(32, 8, 4), (64, 4, 2), (64, 3, 1)], **conv_kwargs):
     '''
 
     def network_fn(X):
-        out = tf.cast(X, tf.float32) / 255.
+        out = X
         with tf.variable_scope("convnet"):
             for num_outputs, kernel_size, stride in convs:
                 out = layers.convolution2d(out,

baselines/common/tests/test_fixed_sequence.py

@@ -6,7 +6,8 @@ from baselines.run import get_learn_function
 
 common_kwargs = dict(
     seed=0,
-    total_timesteps=50000,
+    total_timesteps=20000,
+    nlstm=64
 )
     
 learn_kwargs = {
@@ -19,7 +20,7 @@ learn_kwargs = {
 
 
 alg_list = learn_kwargs.keys()
-rnn_list = ['lstm']
+rnn_list = ['lstm', 'tflstm', 'tflstm_static']
 
 @pytest.mark.slow
 @pytest.mark.parametrize("alg", alg_list)
@@ -41,11 +42,11 @@ def test_fixed_sequence(alg, rnn):
         **kwargs
     )
 
-    simple_test(env_fn, learn, 0.7)
+    simple_test(env_fn, learn, 0.3)
 
 
 if __name__ == '__main__':
-    test_fixed_sequence('ppo2', 'lstm')
+    test_fixed_sequence('ppo2', 'tflstm')
 
     
 

baselines/common/tests/util.py

@@ -2,6 +2,7 @@ import tensorflow as tf
 import numpy as np
 from gym.spaces import np_random
 from baselines.common.vec_env.dummy_vec_env import DummyVecEnv
+from baselines.bench.monitor import Monitor
 
 N_TRIALS = 10000
 N_EPISODES = 100
@@ -10,7 +11,7 @@ def simple_test(env_fn, learn_fn, min_reward_fraction, n_trials=N_TRIALS):
     np.random.seed(0)
     np_random.seed(0)
 
-    env = DummyVecEnv([env_fn])
+    env = DummyVecEnv([lambda: Monitor(env_fn(), None, allow_early_resets=True)])
 
 
     with tf.Graph().as_default(), tf.Session(config=tf.ConfigProto(allow_soft_placement=True)).as_default():

setup.py

@@ -25,8 +25,7 @@ setup(name='baselines',
       extras_require={
         'test': [
             'filelock',
-            'pytest',
-            'pytest-cov',
+            'pytest'
         ]
       },
       description='OpenAI baselines: high quality implementations of reinforcement learning algorithms',