[incubator-mxnet] branch master updated: add ctx to begin_state in rnn_layer (#7580)

[jx...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

jxie pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git


The following commit(s) were added to refs/heads/master by this push:
     new b0b4664  add ctx to begin_state in rnn_layer (#7580)
b0b4664 is described below

commit b0b46641a3d3b58f7b38d8bf84b9bf0c392f3873
Author: Sheng Zha <sz...@users.noreply.github.com>
AuthorDate: Thu Aug 24 10:43:52 2017 -0700

    add ctx to begin_state in rnn_layer (#7580)
    
    * add ctx to begin_state
    
    * fix image classification
---
 example/gluon/image_classification.py | 8 ++++++--
 python/mxnet/gluon/rnn/rnn_layer.py   | 4 ++--
 2 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/example/gluon/image_classification.py b/example/gluon/image_classification.py
index 3f84ff8..b8d018d 100644
--- a/example/gluon/image_classification.py
+++ b/example/gluon/image_classification.py
@@ -57,6 +57,8 @@ parser.add_argument('--batch-norm', action='store_true',
                     help='enable batch normalization or not in vgg. default is false.')
 parser.add_argument('--use-pretrained', action='store_true',
                     help='enable using pretrained model from gluon.')
+parser.add_argument('--kvstore', type=str, default='device',
+                    help='kvstore to use for trainer/module.')
 parser.add_argument('--log-interval', type=int, default=50, help='Number of batches to wait before logging.')
 opt = parser.parse_args()
 
@@ -116,7 +118,8 @@ def train(epochs, ctx):
     if isinstance(ctx, mx.Context):
         ctx = [ctx]
     net.initialize(mx.init.Xavier(magnitude=2.24), ctx=ctx)
-    trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': opt.lr, 'wd': opt.wd})
+    trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': opt.lr, 'wd': opt.wd},
+                            kvstore = opt.kvstore)
     metric = mx.metric.Accuracy()
     loss = gluon.loss.SoftmaxCrossEntropyLoss()
 
@@ -162,7 +165,8 @@ if __name__ == '__main__':
         out = net(data)
         softmax = mx.sym.SoftmaxOutput(out, name='softmax')
         mod = mx.mod.Module(softmax, context=[mx.gpu(i) for i in range(gpus)] if gpus > 0 else [mx.cpu()])
-        mod.fit(train_data, num_epoch=opt.epochs, batch_end_callback = mx.callback.Speedometer(batch_size, 1))
+        mod.fit(train_data, num_epoch=opt.epochs, kvstore=opt.kvstore,
+                batch_end_callback = mx.callback.Speedometer(batch_size, 1))
     else:
         if opt.mode == 'hybrid':
             net.hybridize()
diff --git a/python/mxnet/gluon/rnn/rnn_layer.py b/python/mxnet/gluon/rnn/rnn_layer.py
index 86b7c61..063d566 100644
--- a/python/mxnet/gluon/rnn/rnn_layer.py
+++ b/python/mxnet/gluon/rnn/rnn_layer.py
@@ -141,7 +141,7 @@ class _RNNLayer(Block):
         batch_size: int
             Only required for `NDArray` API. Size of the batch ('N' in layout).
             Dimension of the input.
-        func : callable, default `symbol.zeros`
+        func : callable, default `ndarray.zeros`
             Function for creating initial state.
 
             For Symbol API, func can be `symbol.zeros`, `symbol.uniform`,
@@ -172,7 +172,7 @@ class _RNNLayer(Block):
         batch_size = inputs.shape[self._layout.find('N')]
         skip_states = states is None
         if skip_states:
-            states = self.begin_state(batch_size)
+            states = self.begin_state(batch_size, ctx=inputs.context)
         if isinstance(states, ndarray.NDArray):
             states = [states]
         for state, info in zip(states, self.state_info(batch_size)):

-- 
To stop receiving notification emails like this one, please contact
['"commits@mxnet.apache.org" <co...@mxnet.apache.org>'].