Merge pull request #605 from prabhatnagarajan/fix_tests

Fixes PFN CI tests

.pfnci/run.sh

@@ -24,6 +24,7 @@ TARGET="$1"
 : "${GPU:=0}"
 : "${XPYTEST_NUM_THREADS:=$(nproc)}"
 : "${PYTHON=python3}"
+: "${PYTHON_VERSION=6}"
 : "${CHAINER=}"
 : "${SLOW:=0}"
 
@@ -59,18 +60,25 @@ main() {
       --hint="/chainerrl/.pfnci/hint.pbtxt"
   )
 
+  apt-get update -q -y
+  apt-get install -qy --no-install-recommends software-properties-common
   UBUNTU_VERSION_ID=$(grep DISTRIB_RELEASE /etc/lsb-release | cut -d "=" -f2)
   if [ "$UBUNTU_VERSION_ID" = "16.04" ]; then
     # Because ffmpeg of ubuntu 16.04 causes segmentation fault,
     # we use jonathonf/ffmpeg-3
     apt-get update -q
-    apt-get install -qy --no-install-recommends software-properties-common
     add-apt-repository ppa:cran/ffmpeg-3
   fi
 
-  apt-get update -q
-  apt-get install -qy --no-install-recommends \
-      "${PYTHON}-dev" "${PYTHON}-pip" "${PYTHON}-setuptools" \
+  add-apt-repository ppa:deadsnakes/ppa
+  apt-get update -q -y
+  apt install -y "${PYTHON}.${PYTHON_VERSION}"
+
+
+  apt-get install -qy --no-install-recommends "${PYTHON}.${PYTHON_VERSION}-dev" 
+  apt-get install -qy --no-install-recommends "${PYTHON}-pip"
+  pip3 install --upgrade pip
+  apt-get install -qy --no-install-recommends "${PYTHON}-setuptools" \
       zlib1g-dev make cmake g++ git ffmpeg freeglut3-dev xvfb
 
   if [ "${CHAINER}" != '' ]; then

.pfnci/script.sh

@@ -53,7 +53,6 @@ main() {
     # TODO(imos): Use pre-build Docker images instead because latest images are
     # not updated frequently.
     py3.cpu | py3.gpu ) docker_image=chainer/chainer:latest-python3;;
-    py2.cpu ) docker_image=chainer/chainer:latest-python2;;
     # Unsupported targets.
     * )
       echo "Unsupported target: ${TARGET}" >&2
@@ -63,9 +62,6 @@ main() {
   docker_args+=(--env="CHAINER=${chainer_version}")
   docker_args+=(--env="SLOW=${SLOW:-0}")
 
-  case "${TARGET}" in
-    py2.* ) docker_args+=(--env="PYTHON=python");;
-  esac
 
   for ZIP in a3c_results.zip dqn_results.zip iqn_results.zip rainbow_results.zip ddpg_results.zip trpo_results.zip ppo_results.zip td3_results.zip sac_results.zip
   do

.travis.yml

@@ -36,7 +36,7 @@ install:
 script:
   - flake8 chainerrl
   - flake8 tests
-  - flake8 examples
+  - flake8 --ignore=E402,W504 examples
   - autopep8 -r chainerrl tests examples --diff | tee check_autopep8
   - test ! -s check_autopep8
   - pytest -m "not gpu and not slow" -x tests --cov=chainerrl --ignore tests/misc_tests/test_pretrained_models.py

README.md

@@ -15,7 +15,7 @@ ChainerRL is a deep reinforcement learning library that implements various state
 
 ## Installation
 
-ChainerRL is tested with 3.5.1+. For other requirements, see [requirements.txt](requirements.txt).
+ChainerRL is tested with 3.6. For other requirements, see [requirements.txt](requirements.txt).
 
 ChainerRL can be installed via PyPI:
 ```

chainerrl/functions/sum_arrays.py

@@ -20,7 +20,6 @@ def backward(self, inputs, grads):
         return [grads[0]] * len(inputs)
 
     def forward_gpu(self, inputs):
-        n = len(inputs)
         ptrs = cuda.cupy.asarray([x.data.ptr for x in inputs],
                                  dtype=cuda.cupy.int64)
         y = cuda.elementwise(
@@ -31,7 +30,7 @@ def forward_gpu(self, inputs):
             'for (size_t j = 0; j < n_xs; ++j) {'
             '  y += xs_[j][i];'
             '}',
-            'sum_arrays'.format(n))(inputs[0], ptrs.data.ptr, len(ptrs))
+            'sum_arrays')(inputs[0], ptrs.data.ptr, len(ptrs))
         return y,
 
 

chainerrl/functions/weighted_sum_arrays.py

@@ -23,7 +23,6 @@ def backward(self, inputs, grads):
         return [w * grads[0] for w in self.weights]
 
     def forward_gpu(self, inputs):
-        n = len(inputs)
         ptrs = cuda.cupy.asarray([x.data.ptr for x in inputs],
                                  dtype=cuda.cupy.int64)
         ws = cuda.cupy.asarray(self.weights, dtype=cuda.cupy.float32)
@@ -35,10 +34,10 @@ def forward_gpu(self, inputs):
             'for (size_t j = 0; j < n_xs; ++j) {'
             '  y += xs_[j][i] * ws[j];'
             '}',
-            'weighted_sum_arrays'.format(n))(inputs[0],
-                                             ptrs.data.ptr,
-                                             ws,
-                                             len(ptrs))
+            'weighted_sum_arrays')(inputs[0],
+                                   ptrs.data.ptr,
+                                   ws,
+                                   len(ptrs))
         return y,
 
 

chainerrl/links/mlp.py

@@ -31,6 +31,6 @@ def __init__(self, in_size, out_size, hidden_sizes, nonlinearity=F.relu,
     def __call__(self, x):
         h = x
         if self.hidden_sizes:
-            for l in self.hidden_layers:
-                h = self.nonlinearity(l(h))
+            for link in self.hidden_layers:
+                h = self.nonlinearity(link(h))
         return self.output(h)

chainerrl/links/mlp_bn.py

@@ -74,8 +74,8 @@ def __call__(self, x):
         if self.normalize_input:
             h = self.input_bn(h)
         if self.hidden_sizes:
-            for l in self.hidden_layers:
-                h = self.nonlinearity(l(h))
+            for link in self.hidden_layers:
+                h = self.nonlinearity(link(h))
         h = self.output(h)
         if self.normalize_output:
             h = self.output_bn(h)

chainerrl/misc/init_like_torch.py

@@ -5,16 +5,16 @@
 def init_like_torch(link):
     # Mimic torch's default parameter initialization
     # TODO(muupan): Use chainer's initializers when it is merged
-    for l in link.links():
-        if isinstance(l, L.Linear):
-            out_channels, in_channels = l.W.shape
+    for li in link.links():
+        if isinstance(li, L.Linear):
+            out_channels, in_channels = li.W.shape
             stdv = 1 / np.sqrt(in_channels)
-            l.W.array[:] = np.random.uniform(-stdv, stdv, size=l.W.shape)
-            if l.b is not None:
-                l.b.array[:] = np.random.uniform(-stdv, stdv, size=l.b.shape)
-        elif isinstance(l, L.Convolution2D):
-            out_channels, in_channels, kh, kw = l.W.shape
+            li.W.array[:] = np.random.uniform(-stdv, stdv, size=li.W.shape)
+            if li.b is not None:
+                li.b.array[:] = np.random.uniform(-stdv, stdv, size=li.b.shape)
+        elif isinstance(li, L.Convolution2D):
+            out_channels, in_channels, kh, kw = li.W.shape
             stdv = 1 / np.sqrt(in_channels * kh * kw)
-            l.W.array[:] = np.random.uniform(-stdv, stdv, size=l.W.shape)
-            if l.b is not None:
-                l.b.array[:] = np.random.uniform(-stdv, stdv, size=l.b.shape)
+            li.W.array[:] = np.random.uniform(-stdv, stdv, size=li.W.shape)
+            if li.b is not None:
+                li.b.array[:] = np.random.uniform(-stdv, stdv, size=li.b.shape)

chainerrl/q_functions/dueling_dqn.py

@@ -34,8 +34,8 @@ def __init__(self, n_actions, n_input_channels=4,
 
     def __call__(self, x):
         h = x
-        for l in self.conv_layers:
-            h = self.activation(l(h))
+        for link in self.conv_layers:
+            h = self.activation(link(h))
 
         # Advantage
         batch_size = x.shape[0]
@@ -88,8 +88,8 @@ def __init__(self, n_actions, n_atoms, v_min, v_max,
 
     def __call__(self, x):
         h = x
-        for l in self.conv_layers:
-            h = self.activation(l(h))
+        for link in self.conv_layers:
+            h = self.activation(link(h))
 
         # Advantage
         batch_size = x.shape[0]

chainerrl/recurrent.py

@@ -91,52 +91,52 @@ def state_kept(self):
 def get_state(chain):
     assert isinstance(chain, (chainer.Chain, chainer.ChainList))
     state = []
-    for l in chain.children():
-        if isinstance(l, chainer.links.LSTM):
-            state.append((l.c, l.h))
-        elif isinstance(l, Recurrent):
-            state.append(l.get_state())
-        elif isinstance(l, (chainer.Chain, chainer.ChainList)):
-            state.append(get_state(l))
+    for link in chain.children():
+        if isinstance(link, chainer.links.LSTM):
+            state.append((link.c, link.h))
+        elif isinstance(link, Recurrent):
+            state.append(link.get_state())
+        elif isinstance(link, (chainer.Chain, chainer.ChainList)):
+            state.append(get_state(link))
         else:
             state.append(None)
     return state
 
 
 def stateful_links(chain):
-    for l in chain.children():
-        if isinstance(l, (chainer.links.LSTM, Recurrent)):
-            yield l
-        elif isinstance(l, (chainer.Chain, chainer.ChainList)):
-            for m in stateful_links(l):
+    for link in chain.children():
+        if isinstance(link, (chainer.links.LSTM, Recurrent)):
+            yield link
+        elif isinstance(link, (chainer.Chain, chainer.ChainList)):
+            for m in stateful_links(link):
                 yield m
 
 
 def set_state(chain, state):
     assert isinstance(chain, (chainer.Chain, chainer.ChainList))
-    for l, s in zip(chain.children(), state):
-        if isinstance(l, chainer.links.LSTM):
+    for link, s in zip(chain.children(), state):
+        if isinstance(link, chainer.links.LSTM):
             c, h = s
             # LSTM.set_state doesn't accept None state
             if c is not None:
-                l.set_state(c, h)
-        elif isinstance(l, Recurrent):
-            l.set_state(s)
-        elif isinstance(l, (chainer.Chain, chainer.ChainList)):
-            set_state(l, s)
+                link.set_state(c, h)
+        elif isinstance(link, Recurrent):
+            link.set_state(s)
+        elif isinstance(link, (chainer.Chain, chainer.ChainList)):
+            set_state(link, s)
         else:
             assert s is None
 
 
 def reset_state(chain):
     assert isinstance(chain, (chainer.Chain, chainer.ChainList))
-    for l in chain.children():
-        if isinstance(l, chainer.links.LSTM):
-            l.reset_state()
-        elif isinstance(l, Recurrent):
-            l.reset_state()
-        elif isinstance(l, (chainer.Chain, chainer.ChainList)):
-            reset_state(l)
+    for link in chain.children():
+        if isinstance(link, chainer.links.LSTM):
+            link.reset_state()
+        elif isinstance(link, Recurrent):
+            link.reset_state()
+        elif isinstance(link, (chainer.Chain, chainer.ChainList)):
+            reset_state(link)
 
 
 class RecurrentChainMixin(Recurrent):

docs/install.rst

@@ -5,7 +5,7 @@ Installation
 How to install ChainerRL
 ========================
 
-ChainerRL is tested with 3.5.1+. For other requirements, see ``requirements.txt``.
+ChainerRL is tested with 3.6. For other requirements, see ``requirements.txt``.
 
 .. literalinclude:: ../requirements.txt
   :caption: requirements.txt

examples/atlas/train_soft_actor_critic_atlas.py

@@ -33,7 +33,7 @@ def make_env(args, seed, test):
             raise RuntimeError('roboschool does not work with gym>=0.15.6')
         import roboschool  # NOQA
     env = gym.make(args.env)
-    # Unwrap TimiLimit wrapper
+    # Unwrap TimeLimit wrapper
     assert isinstance(env, gym.wrappers.TimeLimit)
     env = env.env
     # Use different random seeds for train and test envs

examples/grasping/train_dqn_batch_grasping.py

@@ -206,7 +206,6 @@ def make_env(idx, test):
             maxSteps=max_episode_steps,
             isTest=test,
         )
-        assert env.observation_space is None
         env.observation_space = gym.spaces.Box(
             low=0, high=255, shape=(84, 84, 3), dtype=np.uint8)
         # (84, 84, 3) -> (3, 84, 84)