Skip to content

Commit 703affc

Browse files
hycishycis
committed
remove pkg_info
1 parent 5017c30 commit 703affc

File tree

4 files changed

+37
-20
lines changed

4 files changed

+37
-20
lines changed

‎MANIFEST.in‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,2 @@
11
include README.md LICENCE
2-
include ./tensorgraph/pkg_info.json
32
recursive-include tensorgraph *.py

‎setup.py‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,10 @@
11
from distutils.core import setup
22
from setuptools import find_packages
33
import json
4-
5-
with open('./tensorgraph/pkg_info.json') as fp:
6-
_info = json.load(fp)
7-
__version__ = _info['version']
4+
from tensorgraph import __version__
5+
# with open('./tensorgraph/pkg_info.json') as fp:
6+
# _info = json.load(fp)
7+
# __version__ = _info['version']
88

99
setup(
1010
name='tensorgraph',

‎tensorgraph/__init__.py‎

Lines changed: 7 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,11 @@
1-
import json
2-
from os.path import dirname
1+
# import json
2+
# from os.path import dirname
3+
#
4+
# with open(dirname(__file__) + '/pkg_info.json') as fp:
5+
# _info = json.load(fp)
6+
# __version__ = _info['version']
37

4-
with open(dirname(__file__) + '/pkg_info.json') as fp:
5-
_info = json.load(fp)
6-
__version__ = _info['version']
8+
__version = "4.4.10"
79

810
from .stopper import EarlyStopper
911
from .sequential import Sequential

‎tensorgraph/layers/recurrent.py‎

Lines changed: 26 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -6,21 +6,24 @@
66

77
class DynamicLSTM(Template):
88

9-
def __init__(self, num_units, state_is_tuple=True, scope=None):
9+
def __init__(self, num_units, const_seq_len=False, state_is_tuple=True, scope=None):
1010
'''
1111
DESCRIPTION:
1212
DynamicLSTM is for sequences with dynamic length.
1313
PARAMS:
1414
scope (str): scope for the cells. For RNN with the same scope name,
1515
the rnn cell will be reused.
16+
const_seq_len (bool): if true, will use a constant sequence
17+
length equal to the max_time. state_below will just be
18+
singular input instead of tuple.
1619
'''
1720
if scope is None:
1821
self.scope = self.__class__.__name__
1922
else:
2023
self.scope = scope
2124
with tf.variable_scope(self.scope):
2225
self.lstm = tf.contrib.rnn.LSTMCell(num_units=num_units, state_is_tuple=state_is_tuple)
23-
26+
self.const_seq_len = const_seq_len
2427

2528

2629
def _train_fprop(self, state_below):
@@ -36,9 +39,12 @@ def _train_fprop(self, state_below):
3639
if state_is_tuple = False:
3740
return tf.concat(1, [C, h]) of dimension [batchsize, 2 * num_units]
3841
'''
39-
X_sb, seqlen_sb = state_below
40-
with tf.variable_scope(self.scope) as scope:
42+
if self.const_seq_len:
43+
X_sb, seqlen_sb = state_below, None
44+
else:
45+
X_sb, seqlen_sb = state_below
4146

47+
with tf.variable_scope(self.scope) as scope:
4248
try:
4349
bef = set(tf.global_variables())
4450
outputs, last_states = tf.nn.dynamic_rnn(cell=self.lstm,
@@ -63,7 +69,7 @@ def _variables(self):
6369

6470
class LSTM(Template):
6571

66-
def __init__(self, num_units, return_idx, initial_state=None, state_is_tuple=True, scope=None):
72+
def __init__(self, num_units, return_idx=[0,1,2], initial_state=None, state_is_tuple=True, scope=None):
6773
'''
6874
DESCRIPTION:
6975
LSTM is for sequences with fixed length.
@@ -75,7 +81,9 @@ def __init__(self, num_units, return_idx, initial_state=None, state_is_tuple=Tru
7581
[batch_size, cell.state_size]. If cell.state_size is a tuple,
7682
this should be a tuple of tensors having shapes [batch_size, s]
7783
for s in cell.state_size.
78-
return_idx (list): list of index from the rnn outputs to return
84+
return_idx (list): list of index from the rnn outputs to return from
85+
[outputs, context, last_hid], indexes has to fall into
86+
[0, 1, 2]
7987
'''
8088
if scope is None:
8189
self.scope = self.__class__.__name__
@@ -86,11 +94,11 @@ def __init__(self, num_units, return_idx, initial_state=None, state_is_tuple=Tru
8694
self.initial_state = initial_state
8795
self.state_is_tuple = state_is_tuple
8896
self.return_idx = return_idx
89-
assert max(self.return_idx) <= 2 and min(self.return_idx) >= 0
97+
assert max(self.return_idx) <= 2 and min(self.return_idx) >= 0, 'indexes \
98+
does not fall into [outputs, context, last_hid]'
9099
assert isinstance(self.return_idx, list)
91100

92101

93-
94102
def _train_fprop(self, state_below):
95103
'''
96104
PARAMS:
@@ -135,13 +143,16 @@ def _variables(self):
135143

136144
class DynamicBiLSTM(Template):
137145

138-
def __init__(self, fw_num_units, bw_num_units, state_is_tuple=True, scope=None):
146+
def __init__(self, fw_num_units, bw_num_units, const_seq_len=False, state_is_tuple=True, scope=None):
139147
'''
140148
DESCRIPTION:
141149
BiDynamicLSTM is for sequences with dynamic length.
142150
PARAMS:
143151
scope (str): scope for the cells. For RNN with the same scope name,
144152
the rnn cell will be reused.
153+
const_seq_len (bool): if true, will use a constant sequence
154+
length equal to the max_time. state_below will just be
155+
singular input instead of tuple.
145156
'''
146157
if scope is None:
147158
self.scope = self.__class__.__name__
@@ -150,6 +161,7 @@ def __init__(self, fw_num_units, bw_num_units, state_is_tuple=True, scope=None):
150161
with tf.variable_scope(self.scope):
151162
self.fw_lstm = tf.contrib.rnn.LSTMCell(num_units=fw_num_units, state_is_tuple=state_is_tuple)
152163
self.bw_lstm = tf.contrib.rnn.LSTMCell(num_units=bw_num_units, state_is_tuple=state_is_tuple)
164+
self.const_seq_len = const_seq_len
153165

154166

155167
def _train_fprop(self, state_below):
@@ -167,7 +179,11 @@ def _train_fprop(self, state_below):
167179
if state_is_tuple = False:
168180
return tf.concat(1, [C, h]) of dimension [batchsize, 2 * num_units]
169181
'''
170-
X_sb, seqlen_sb = state_below
182+
if self.const_seq_len:
183+
seqlen_sb = None
184+
X_sb = state_below
185+
else:
186+
X_sb, seqlen_sb = state_below
171187

172188
with tf.variable_scope(self.scope) as scope:
173189
try:

0 commit comments

Comments
 (0)