Skip to content
This repository has been archived by the owner on Aug 14, 2019. It is now read-only.

Commit

Permalink
Upgrade to Tensorflow 1.1.0 (#32)
Browse files Browse the repository at this point in the history 
* Update TF version in requirements.txt

* Update switchable dropout test for tf 1.1.0

* Update RNNcell behavior to match tf 1.1.0 expectations

* Update SiameseMatchingBiLSTM for new tf 1.1.0 RNNCell behavior

* Remove print statement from switchable dropout test

* Fix lint

* Log the proper dropout LSTM output

* Fix dropout wrapper test and remove print

* Fix lint
  • Loading branch information
@nelson-liu
nelson-liu committed Apr 29, 2017
1 parent b868570 commit d26d393
Show file tree
Hide file tree
Showing 4 changed files with 75 additions and 46 deletions.
34 changes: 24 additions & 10 deletions duplicate_questions/models/siamese_bilstm/siamese_bilstm.py
View file
Expand Up @@ -173,15 +173,19 @@ def _build_forward(self):
rnn_hidden_size = self.rnn_hidden_size
rnn_output_mode = self.rnn_output_mode
output_keep_prob = self.output_keep_prob
rnn_cell = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell = SwitchableDropoutWrapper(rnn_cell,
self.is_train,
output_keep_prob=output_keep_prob)
rnn_cell_fw_one = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell_fw_one = SwitchableDropoutWrapper(rnn_cell_fw_one,
self.is_train,
output_keep_prob=output_keep_prob)
rnn_cell_bw_one = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell_bw_one = SwitchableDropoutWrapper(rnn_cell_bw_one,
self.is_train,
output_keep_prob=output_keep_prob)
with tf.variable_scope("encode_sentences"):
# Encode the first sentence.
(fw_output_one, bw_output_one), _ = tf.nn.bidirectional_dynamic_rnn(
cell_fw=d_rnn_cell,
cell_bw=d_rnn_cell,
cell_fw=d_rnn_cell_fw_one,
cell_bw=d_rnn_cell_bw_one,
dtype="float",
sequence_length=sentence_one_len,
inputs=word_embedded_sentence_one,
Expand All @@ -190,17 +194,27 @@ def _build_forward(self):
# Encode the second sentence, using the same RNN weights.
tf.get_variable_scope().reuse_variables()
(fw_output_two, bw_output_two), _ = tf.nn.bidirectional_dynamic_rnn(
cell_fw=d_rnn_cell,
cell_bw=d_rnn_cell,
cell_fw=d_rnn_cell_fw_one,
cell_bw=d_rnn_cell_bw_one,
dtype="float",
sequence_length=sentence_two_len,
inputs=word_embedded_sentence_two,
scope="encoded_sentence_one")
else:
# Encode the second sentence with a different RNN
rnn_cell_fw_two = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell_fw_two = SwitchableDropoutWrapper(
rnn_cell_fw_two,
self.is_train,
output_keep_prob=output_keep_prob)
rnn_cell_bw_two = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell_bw_two = SwitchableDropoutWrapper(
rnn_cell_bw_two,
self.is_train,
output_keep_prob=output_keep_prob)
(fw_output_two, bw_output_two), _ = tf.nn.bidirectional_dynamic_rnn(
cell_fw=d_rnn_cell,
cell_bw=d_rnn_cell,
cell_fw=d_rnn_cell_fw_two,
cell_bw=d_rnn_cell_bw_two,
dtype="float",
sequence_length=sentence_two_len,
inputs=word_embedded_sentence_two,
Expand Down
34 changes: 24 additions & 10 deletions duplicate_questions/models/siamese_bilstm/siamese_matching_bilstm.py
View file
Expand Up @@ -171,15 +171,19 @@ def _build_forward(self):

rnn_hidden_size = self.rnn_hidden_size
output_keep_prob = self.output_keep_prob
rnn_cell = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell = SwitchableDropoutWrapper(rnn_cell,
self.is_train,
output_keep_prob=output_keep_prob)
rnn_cell_fw_one = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell_fw_one = SwitchableDropoutWrapper(rnn_cell_fw_one,
self.is_train,
output_keep_prob=output_keep_prob)
rnn_cell_bw_one = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell_bw_one = SwitchableDropoutWrapper(rnn_cell_bw_one,
self.is_train,
output_keep_prob=output_keep_prob)
with tf.variable_scope("encode_sentences"):
# Encode the first sentence.
(fw_output_one, bw_output_one), _ = tf.nn.bidirectional_dynamic_rnn(
cell_fw=d_rnn_cell,
cell_bw=d_rnn_cell,
cell_fw=d_rnn_cell_fw_one,
cell_bw=d_rnn_cell_bw_one,
dtype="float",
sequence_length=sentence_one_len,
inputs=word_embedded_sentence_one,
Expand All @@ -188,17 +192,27 @@ def _build_forward(self):
# Encode the second sentence, using the same RNN weights.
tf.get_variable_scope().reuse_variables()
(fw_output_two, bw_output_two), _ = tf.nn.bidirectional_dynamic_rnn(
cell_fw=d_rnn_cell,
cell_bw=d_rnn_cell,
cell_fw=d_rnn_cell_fw_one,
cell_bw=d_rnn_cell_bw_one,
dtype="float",
sequence_length=sentence_two_len,
inputs=word_embedded_sentence_two,
scope="encoded_sentence_one")
else:
# Encode the second sentence with a different RNN
rnn_cell_fw_two = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell_fw_two = SwitchableDropoutWrapper(
rnn_cell_fw_two,
self.is_train,
output_keep_prob=output_keep_prob)
rnn_cell_bw_two = LSTMCell(rnn_hidden_size, state_is_tuple=True)
d_rnn_cell_bw_two = SwitchableDropoutWrapper(
rnn_cell_bw_two,
self.is_train,
output_keep_prob=output_keep_prob)
(fw_output_two, bw_output_two), _ = tf.nn.bidirectional_dynamic_rnn(
cell_fw=d_rnn_cell,
cell_bw=d_rnn_cell,
cell_fw=d_rnn_cell_fw_two,
cell_bw=d_rnn_cell_bw_two,
dtype="float",
sequence_length=sentence_two_len,
inputs=word_embedded_sentence_two,
Expand Down
2 changes: 1 addition & 1 deletion requirements.txt
View file
@@ -1,7 +1,7 @@
#### ESSENTIAL LIBRARIES FOR MAIN FUNCTIONALITY ####

# Neural network and related libraries
tensorflow==1.0.1
tensorflow==1.1.0
numpy

# Progress logger
Expand Down
51 changes: 26 additions & 25 deletions tests/util/test_switchable_dropout_wrapper.py
View file
Expand Up @@ -36,28 +36,29 @@ def test_switchable_dropout_wrapper_state_is_tuple(self):
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
output_no_train = rnn_output.eval(feed_dict={is_train: False})
expected_output_no_train = np.array([[[0.08825343, 0.01838959, 0.01872513],
[0.01195384, -0.14495267, 0.31236988],
expected_output_no_train = np.array([[[0.10523333, -0.03578992, 0.16407447],
[-0.07642615, -0.1346959, 0.07218226],
[0.0, 0.0, 0.0]],
[[0.0143504, -0.14128588, 0.07712727],
[[-0.31979755, -0.12604457, -0.24436688],
[0.0, 0.0, 0.0],
[0.0, 0.0, 0.0]],
[[-0.10839351, 0.05113239, -0.04910426],
[-0.06215987, 0.16528113, -0.00543074],
[0.05128077, 0.23328263, -0.04104931]]])
[[0.27140033, -0.01063369, 0.11808267],
[0.15138564, -0.10808259, 0.13118345],
[0.20397078, -0.06317351, 0.21408504]]])
assert_allclose(output_no_train, expected_output_no_train, rtol=1e-06)

output_train = rnn_output.eval(feed_dict={is_train: True})
expected_output_train = np.array([[[0.0, 0.03697259, 0.0],
[-0.52872497, 0.03966674, -0.24180387],
expected_output_train = np.array([[[-0.0, -0.21935862, -0.11160457],
[-0.0, -0.0, 0.09479073],
[0.0, 0.0, 0.0]],
[[0.24703449, -0.06437484, 0.15770586],
[[0.02565068, 0.21709232, -0.0],
[0.0, 0.0, 0.0],
[0.0, 0.0, 0.0]],
[[-0.23952101, -0.03936104, 0.0442652],
[-0.3688741, -0.08598197, 0.0],
[-0.00903562, 0.0, -0.03707428]]])
[[0.0, 0.0, 0.07740743],
[0.04682902, -0.14770079, 0.14597748],
[0.0, 0.09399685, 0.0]]])
# low precision test, this one seems flaky
assert_allclose(output_train, expected_output_train, rtol=1e-04)
assert_allclose(output_train, expected_output_train, rtol=1e-06)

@flaky
def test_switchable_dropout_wrapper_state_is_not_tuple(self):
Expand Down Expand Up @@ -87,24 +88,24 @@ def test_switchable_dropout_wrapper_state_is_not_tuple(self):
sess.run(tf.global_variables_initializer())
output_no_train = rnn_output.eval(feed_dict={is_train: False})
expected_output_no_train = np.array(
[[[-0.09445292, -0.08269257, 0.1921162],
[0.13077924, -0.16224632, 0.07092731],
[[[-0.10366952, -0.01751264, -0.02237115],
[-0.07636562, 0.06660741, 0.02946584],
[0.0, 0.0, 0.0]],
[[0.15248382, 0.18584363, -0.12413846],
[[-0.09134783, 0.15928121, 0.05786164],
[0.0, 0.0, 0.0],
[0.0, 0.0, 0.0]],
[[-0.07077862, -0.11894269, 0.33125928],
[-0.17721169, -0.07561724, 0.25253388],
[-0.18928067, -0.07377248, 0.41105911]]])
[[-0.00575439, -0.22505699, -0.27295753],
[-0.12970942, -0.16395324, -0.06502352],
[-0.16302694, -0.27601245, -0.20045257]]])
assert_allclose(output_no_train, expected_output_no_train, rtol=1e-06)
output_train = rnn_output.eval(feed_dict={is_train: True})
expected_output_train = np.array([[[0.0, 0.04103347, -0.0],
[0.26753482, -0.03276764, 0.0240659],
expected_output_train = np.array([[[-0.0, 0.13120674, -0.02568678],
[-0.0, 0.0, -0.20105337],
[0.0, 0.0, 0.0]],
[[-0.11665035, -0.00709306, 0.01923252],
[[-0.02063255, 0.25306353, 0.0],
[0.0, 0.0, 0.0],
[0.0, 0.0, 0.0]],
[[0.04758045, -0.03102016, -0.04817296],
[0.18634762, -0.04973229, 0.0],
[-0.10404891, 0.0, 0.0632768]]])
[[0.0, -0.0, -0.0595048],
[0.03207482, -0.07930075, -0.09382694],
[0.0, -0.00405498, -0.0]]])
assert_allclose(output_train, expected_output_train, rtol=1e-06)

0 comments on commit d26d393

Please sign in to comment.