def testArithmeticRenames(self):
with self.test_session() as s:
stuff = tf.split(1, 2, [[1, 2, 3, 4], [4, 5, 6, 7]])
vals = s.run(stuff)
self.assertAllEqual(vals, [[[1, 2], [4, 5]], [[3, 4], [6, 7]]])
self.assertAllEqual(tf.neg(tf.mul(tf.add(1, 2), tf.sub(5, 3))).eval(), -6)
self.assertAllEqual(s.run(tf.listdiff([1, 2, 3], [3, 3, 4]))[0], [1, 2])
self.assertAllEqual(tf.list_diff([1, 2, 3], [3, 3, 4])[0].eval(), [1, 2])
a = [[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]
foo = np.where(np.less(a, 2), np.negative(a), a)
self.assertAllEqual(tf.select(tf.less(a, 2), tf.neg(a), a).eval(), foo)
self.assertAllEqual(tf.complex_abs(tf.constant(3 + 4.0j)).eval(), 5)
def _testListDiff(self, x, y, out, idx, dtype=np.int32):
x = np.array(x, dtype=dtype)
y = np.array(y, dtype=dtype)
out = np.array(out, dtype=dtype)
idx = np.array(idx, dtype=dtype)
with self.test_session() as sess:
x_tensor = tf.convert_to_tensor(x)
y_tensor = tf.convert_to_tensor(y)
out_tensor, idx_tensor = tf.listdiff(x_tensor, y_tensor)
tf_out, tf_idx = sess.run([out_tensor, idx_tensor])
self.assertAllEqual(tf_out, out)
self.assertAllEqual(tf_idx, idx)
self.assertEqual(1, out_tensor.get_shape().ndims)
self.assertEqual(1, idx_tensor.get_shape().ndims)
def testArithmeticRenames(self):
with self.cached_session() as s:
stuff = tf.split(1, 2, [[1, 2, 3, 4], [4, 5, 6, 7]])
vals = s.run(stuff)
self.assertAllEqual(vals, [[[1, 2], [4, 5]], [[3, 4], [6, 7]]])
self.assertAllEqual(
tf.neg(tf.mul(tf.add(1, 2), tf.sub(5, 3))).eval(), -6)
self.assertAllEqual(
s.run(tf.listdiff([1, 2, 3], [3, 3, 4]))[0], [1, 2])
self.assertAllEqual(
tf.list_diff([1, 2, 3], [3, 3, 4])[0].eval(), [1, 2])
a = [[1., 2., 3.], [4., 5., 6.]]
foo = np.where(np.less(a, 2), np.negative(a), a)
self.assertAllEqual(
tf.select(tf.less(a, 2), tf.neg(a), a).eval(), foo)
self.assertAllEqual(tf.complex_abs(tf.constant(3 + 4.j)).eval(), 5)
def _testListDiff(self, x, y, out, idx, dtype=np.int32):
x = np.array(x, dtype=dtype)
y = np.array(y, dtype=dtype)
out = np.array(out, dtype=dtype)
idx = np.array(idx, dtype=dtype)
with self.test_session() as sess:
x_tensor = tf.convert_to_tensor(x)
y_tensor = tf.convert_to_tensor(y)
out_tensor, idx_tensor = tf.listdiff(x_tensor, y_tensor)
tf_out, tf_idx = sess.run([out_tensor, idx_tensor])
self.assertAllEqual(tf_out, out)
self.assertAllEqual(tf_idx, idx)
self.assertEqual(1, out_tensor.get_shape().ndims)
self.assertEqual(1, idx_tensor.get_shape().ndims)
def _testListDiff(self, x, y, out, idx):
for dtype in _TYPES:
if dtype == tf.string:
x = [tf.compat.as_bytes(str(a)) for a in x]
y = [tf.compat.as_bytes(str(a)) for a in y]
out = [tf.compat.as_bytes(str(a)) for a in out]
with self.test_session() as sess:
x_tensor = tf.convert_to_tensor(x, dtype=dtype)
y_tensor = tf.convert_to_tensor(y, dtype=dtype)
out_tensor, idx_tensor = tf.listdiff(x_tensor, y_tensor)
tf_out, tf_idx = sess.run([out_tensor, idx_tensor])
self.assertAllEqual(tf_out, out)
self.assertAllEqual(tf_idx, idx)
self.assertEqual(1, out_tensor.get_shape().ndims)
self.assertEqual(1, idx_tensor.get_shape().ndims)
def _testListDiff(self, x, y, out, idx):
for dtype in _TYPES:
if dtype == tf.string:
x = [tf.compat.as_bytes(str(a)) for a in x]
y = [tf.compat.as_bytes(str(a)) for a in y]
out = [tf.compat.as_bytes(str(a)) for a in out]
with self.test_session() as sess:
x_tensor = tf.convert_to_tensor(x, dtype=dtype)
y_tensor = tf.convert_to_tensor(y, dtype=dtype)
out_tensor, idx_tensor = tf.listdiff(x_tensor, y_tensor)
tf_out, tf_idx = sess.run([out_tensor, idx_tensor])
self.assertAllEqual(tf_out, out)
self.assertAllEqual(tf_idx, idx)
self.assertEqual(1, out_tensor.get_shape().ndims)
self.assertEqual(1, idx_tensor.get_shape().ndims)
def forward(self, phrase_max_size=1, composition_function='RNN', dim=100, batch_size=1, neg=1,
freq_table=[], init_word_matrix=[], init_context_matrix=[], embedding_train=False):
if len(init_word_matrix) > 0:
embed = tf.Variable(initial_value=init_word_matrix, trainable=embedding_train, name='embed')
else:
embed = tf.Variable(initial_value=tf.random_uniform([len(freq_table), dim], -0.5 / dim, 0.5 / dim),
trainable=embedding_train, name='embed')
self._embed = embed
if len(init_context_matrix) > 0:
context_emb = tf.Variable(initial_value=init_context_matrix, trainable=embedding_train, name='context_emb')
else:
context_emb = tf.Variable(initial_value=tf.zeros([len(freq_table), dim]),
trainable=embedding_train, name='context_emb')
self._context_emb = context_emb
holder, composed = self.construct_composition(phrase_max_size, composition_function, dim, batch_size, embedding_train)
context_id = tf.placeholder(tf.int32, [batch_size])
#negative sampling for each example
sampled_ids, _, _ = (tf.nn.fixed_unigram_candidate_sampler(
true_classes=tf.reshape(tf.cast(context_id, dtype=tf.int64), [batch_size, 1]),
num_true=1,
num_sampled=neg+1,
unique=True,
range_max=len(freq_table),
distortion=0.75,
unigrams=freq_table))
exclude_list = tf.cast(tf.constant([self._word_id['</s>']]), dtype=tf.int64) #exclude terminal node from negative sampling result
sampled_ids, _ = tf.listdiff(sampled_ids, exclude_list)
true_logit = {}
negative_logit = {}
true_context = tf.nn.embedding_lookup(self._context_emb, context_id)
negative_context = tf.nn.embedding_lookup(self._context_emb, sampled_ids[:neg])
for length, compose_embed in composed.iteritems():
#calculate prob for positive example
true_logit[length] = tf.reduce_sum(tf.mul(compose_embed, true_context), 1)
negative_logit[length] = tf.matmul(compose_embed, negative_context, transpose_b=True)
return holder, composed, context_id, true_logit, negative_logit
import tensorflow as tf sess = tf.InteractiveSession() x = tf.constant([[2, 5, 3, -5], [0, 3, -2, 5], [4, 3, 5, 3], [6, 1, 4, 0]]) listx = tf.constant([1, 2, 3, 4, 5, 6, 7, 8]) listy = tf.constant([4, 5, 8, 9]) boolx = tf.constant([[True, False], [False, True]]) tf.argmin(x, 1).eval() # Position of the maximum value of columns tf.argmax(x, 1).eval() # Position of the minimum value of rows tf.listdiff(listx, listy)[0].eval() # List differences tf.where(boolx).eval() # Show true values tf.unique(listx)[0].eval() # Unique values in list
def test_ListDiff(self):
if td._tf_version[:2] <= (0, 11):
l = np.random.randint(0, 5, 100)
t1, t2 = tf.listdiff(l, l[::-2])
self.check(t1)
self.check(t2)
import tensorflow as tf
import numpy as np
# tf.argmin
x = np.random.rand(10, 5, 4)
z_argmin = tf.argmin(x, dimension=0)
# tf.argmax
x = np.random.rand(10, 5, 4)
z_argmax = tf.argmax(x, dimension=1)
# tf.listdiff
x = np.random.randint(0, 10, 100)
y = np.random.randint(0, 10, 10)
z_listdiff = tf.listdiff(x, y)
# tf.where
x = np.random.randint(0, 2, 10 * 5 * 4)
z = np.empty(10 * 5 * 4, dtype=np.bool)
for i, x_ in enumerate(x):
if x_ > 0:
z[i] = True
else:
z[i] = False
z = z.reshape((10, 5, 4))
z_where = tf.where(z)
# tf.unique
x = np.random.randint(0, 10, 100)
z_unique = tf.unique(x)
def test_ListDiff(self):
l = np.random.randint(0, 5, 100)
t1, t2 = tf.listdiff(l, l[::-2])
self.check(t1)
self.check(t2)
import tensorflow as tf
sess = tf.InteractiveSession()
x = tf.constant([[2, 5, 3, -5],
[0, 3,-2, 5],
[4, 3, 5, 3],
[6, 1, 4, 0]])
listx = tf.constant([1,2,3,4,5,6,7,8])
listy = tf.constant([4,5,8,9])
boolx = tf.constant([[True,False], [False,True]])
tf.argmin(x, 1).eval() # Position of the maximum value of columns
tf.argmax(x, 1).eval() # Position of the minimum value of rows
tf.listdiff(listx, listy)[0].eval() # List differences
tf.where(boolx).eval() # Show true values
tf.unique(listx)[0].eval() # Unique values in list
tf.unsorted_segment_sum(data, segment_ids,
num_segments, name=None) 与tf.segment_sum函数类似，
不同在于segment_ids中id顺序可以是无序的
tf.sparse_segment_sum(data, indices,
segment_ids, name=None) 输入进行稀疏分割求和
c = tf.constant([[1,2,3,4], [-1,-2,-3,-4], [5,6,7,8]])
# Select two rows, one segment.
tf.sparse_segment_sum(c, tf.constant([0, 1]), tf.constant([0, 0]))
==> [[0 0 0 0]]
对原data的indices为[0,1]位置的进行分割，
并按照segment_ids的分组进行求和
七、序列比较与索引提取(Sequence Comparison and Indexing)
tf.argmin(input, dimension, name=None) 返回input最小值的索引index
tf.argmax(input, dimension, name=None) 返回input最大值的索引index
tf.listdiff(x, y, name=None) 返回x，y中不同值的索引
tf.where(input, name=None) 返回bool型tensor中为True的位置
# ‘input’ tensor is
#[[True, False]
#[True, False]]
# ‘input’ 有两个’True’,那么输出两个坐标值.
# ‘input’的rank为2, 所以每个坐标为具有两个维度.
where(input) ==>
[[0, 0],
[1, 0]]
tf.unique(x, name=None) 返回一个元组tuple(y,idx)，y为x的列表的唯一化数据列表，
idx为x数据对应y元素的index
# tensor ‘x’ is [1, 1, 2, 4, 4, 4, 7, 8, 8]
y, idx = unique(x)
y ==> [1, 2, 4, 7, 8]
idx ==> [0, 0, 1, 2, 2, 2, 3, 4, 4]
