word2vec_basic.py

ssh://sci@192.168.67.128:22/usr/bin/python3 -u /home/win_pymine_clean/feature_wifi/word2vec_basic.py
Found and verified text8.zip
Data size 17005207
Most common words (+UNK) [[‘UNK‘, 418391], (‘the‘, 1061396), (‘of‘, 593677), (‘and‘, 416629), (‘one‘, 411764)]
Sample data [5238, 3084, 12, 6, 195, 2, 3136, 46, 59, 156] [‘anarchism‘, ‘originated‘, ‘as‘, ‘a‘, ‘term‘, ‘of‘, ‘abuse‘, ‘first‘, ‘used‘, ‘against‘]
3084 originated -> 5238 anarchism
3084 originated -> 12 as
12 as -> 6 a
12 as -> 3084 originated
6 a -> 195 term
6 a -> 12 as
195 term -> 2 of
195 term -> 6 a
2017-09-18 01:35:09.854800: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn‘t compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2017-09-18 01:35:09.874305: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn‘t compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2017-09-18 01:35:09.874359: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn‘t compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2017-09-18 01:35:09.874379: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn‘t compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2017-09-18 01:35:09.874396: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn‘t compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
Initialized
Average loss at step  0 :  261.962097168
Nearest to but: rtp, sheng, modules, defenders, banknote, contrasting, proximal, va,
Nearest to state: pigeon, neutralize, gacy, independent, lug, rebellious, scholar, infestation,
Nearest to five: analytics, bhp, exams, synapomorphies, paolo, nanotubes, leaching, lombards,
Nearest to their: fragment, allocations, filthy, forgets, leaks, myron, albatross, response,
Nearest to as: bodybuilders, expecting, indivisible, rembrandt, clades, redirection, summand, checkmate,
Nearest to other: editorship, aan, beauties, exterminated, guan, nolo, bess, pearce,
Nearest to than: explainable, sword, capacitor, savory, rally, coleridge, imr, subordinates,
Nearest to are: ignatius, repetitive, eras, soter, elan, collectivity, fractions, perugia,
Nearest to its: castillo, hogshead, rna, ljubljana, fiberglass, ausgleich, toxin, balderus,
Nearest to see: abrasive, rifles, und, sufferer, nvidia, caricatured, manly, chee,
Nearest to and: meiosis, bridges, insurrection, feistel, sacrum, sepsis, reinforced, oscillator,
Nearest to three: honest, communally, gast, hawkins, highway, menzies, flocked, nijmegen,
Nearest to it: peh, namespaces, revolting, murat, mcgrath, toltec, fiorello, aediles,
Nearest to war: buprenorphine, rakis, decide, funky, stilicho, rnir, satirized, hydrolysis,
Nearest to however: pyramid, whittier, edifices, heraclea, alveolar, counties, carousel, carnap,
Nearest to states: renormalization, ideology, defending, immerse, semi, fireplace, roach, classically,
Average loss at step  2000 :  113.145377743
Average loss at step  4000 :  53.0118778615
Average loss at step  6000 :  33.3697857151
Average loss at step  8000 :  23.4925061049
Average loss at step  10000 :  17.8379190623
Nearest to but: and, soroban, var, electromagnetism, suitable, yeast, ruins, connotation,
Nearest to state: feel, independent, refrigerator, scholar, inventor, once, archive, disambiguation,
Nearest to five: zero, bckgr, nine, coke, eight, one, seven, hakama,
Nearest to their: response, wiesenthal, the, fragment, a, naturally, dawn, prevent,
Nearest to as: by, namibia, is, in, encourage, and, heroes, nazi,
Nearest to other: clock, concluded, bang, house, western, existent, obey, strategic,
Nearest to than: sword, rally, equilibrium, first, for, dhabi, contained, sample,
Nearest to are: is, under, certain, were, fractions, was, wickets, ignatius,
Nearest to its: faber, the, explanations, aristophanes, rna, ed, apo, one,
Nearest to see: abbot, entire, yeast, kohanim, rifles, solstice, multi, jurisdiction,
Nearest to and: in, of, with, or, nine, the, UNK, by,
Nearest to three: zero, jpg, nine, one, two, six, seven, five,
Nearest to it: archie, ideas, he, decided, viet, sigma, a, this,
Nearest to war: decide, density, from, patent, supposedly, passenger, agave, evil,
Nearest to however: pyramid, and, atheists, cuisine, counties, neq, spotted, technique,
Nearest to states: agave, semi, ideology, what, nine, defending, defendants, classically,
Average loss at step  12000 :  14.0006338406
Average loss at step  14000 :  11.8489933434
Average loss at step  16000 :  9.97434587622
Average loss at step  18000 :  8.41560820138
Average loss at step  20000 :  7.94222211516
Nearest to but: and, circ, electromagnetism, northrop, suitable, soroban, is, which,
Nearest to state: subkey, refrigerator, feel, inventor, once, rebellious, first, independent,
Nearest to five: nine, eight, zero, seven, six, three, four, agouti,
Nearest to their: his, the, acacia, a, wiesenthal, fragment, response, its,
Nearest to as: and, in, for, by, is, agouti, from, harbour,
Nearest to other: concluded, clock, existent, operatorname, western, rai, obey, bang,
Nearest to than: for, sword, rally, archimedean, antwerp, equilibrium, or, nine,
Nearest to are: is, were, was, agouti, have, ignatius, gollancz, under,
Nearest to its: the, his, operatorname, rna, magdeburg, their, faber, requisite,
Nearest to see: kohanim, seabirds, agouti, abbot, entire, yeast, rifles, subkey,
Nearest to and: or, circ, dasyprocta, operatorname, in, agouti, of, for,
Nearest to three: six, two, nine, seven, five, eight, one, zero,
Nearest to it: he, this, she, there, archie, which, viet, dasyprocta,
Nearest to war: decide, density, passenger, patent, operatorname, stilicho, dasyprocta, ns,
Nearest to however: and, cuisine, educators, pyramid, edifices, credo, atheists, operatorname,
Nearest to states: agave, agouti, semi, dasyprocta, ideology, hadith, what, defending,
Average loss at step  22000 :  7.19506471896
Average loss at step  24000 :  6.85889063323
Average loss at step  26000 :  6.72703241718
Average loss at step  28000 :  6.40524325681
Average loss at step  30000 :  5.92727108788
Nearest to but: and, circ, which, suitable, electromagnetism, although, operatorname, agouti,
Nearest to state: subkey, feel, refrigerator, inventor, diodes, rebellious, first, once,
Nearest to five: eight, six, seven, four, three, zero, nine, two,
Nearest to their: his, the, its, acacia, a, wiesenthal, fragment, punts,
Nearest to as: by, harbour, and, agouti, in, circ, is, for,
Nearest to other: clock, concluded, western, rai, different, operatorname, traits, existent,
Nearest to than: or, for, rally, sword, and, ann, archimedean, antwerp,
Nearest to are: were, is, was, have, agouti, be, under, gollancz,
Nearest to its: the, their, his, operatorname, magdeburg, requisite, rna, tonnage,
Nearest to see: kohanim, seabirds, trinomial, agouti, aba, abbot, yeast, subkey,
Nearest to and: or, circ, dasyprocta, agouti, operatorname, with, in, of,
Nearest to three: six, five, two, four, seven, eight, nine, agouti,
Nearest to it: he, this, she, there, which, amalthea, archie, dasyprocta,
Nearest to war: decide, buprenorphine, density, ns, patent, passenger, aforementioned, northrop,
Nearest to however: and, almighty, educators, cuisine, technique, edifices, expressly, operatorname,
Nearest to states: agave, semi, agouti, defending, ideology, dasyprocta, hadith, what,
Average loss at step  32000 :  5.91289714551
Average loss at step  34000 :  5.74882777894
Average loss at step  36000 :  5.79514622164
Average loss at step  38000 :  5.50678780043
Average loss at step  40000 :  5.23573713326
Nearest to but: and, circ, although, which, or, zero, that, electromagnetism,
Nearest to state: subkey, inventor, feel, refrigerator, diodes, rebellious, independent, first,
Nearest to five: four, three, six, eight, seven, zero, two, nine,
Nearest to their: his, its, the, acacia, operatorname, wiesenthal, agouti, punts,
Nearest to as: agouti, harbour, circ, zero, and, in, by, aba,
Nearest to other: different, clock, western, operatorname, concluded, fountain, dasyprocta, rai,
Nearest to than: or, for, rally, and, sword, simple, hbf, qualification,
Nearest to are: were, is, have, agouti, zero, be, was, multivibrator,
Nearest to its: their, the, his, operatorname, magdeburg, dasyprocta, agouti, zero,
Nearest to see: kohanim, trinomial, seabirds, agouti, aba, rifles, abrasive, abbot,
Nearest to and: or, zero, circ, dasyprocta, operatorname, in, agouti, but,
Nearest to three: four, five, six, seven, eight, two, zero, agouti,
Nearest to it: he, she, this, which, there, zero, archie, sal,
Nearest to war: decide, buprenorphine, northrop, density, ordinarily, aforementioned, slovenes, collide,
Nearest to however: and, but, educators, almighty, technique, that, operatorname, cuisine,
Nearest to states: semi, agouti, agave, defending, ideology, dasyprocta, hadith, what,
Average loss at step  42000 :  5.36539960575
Average loss at step  44000 :  5.24343806195
Average loss at step  46000 :  5.22763335347
Average loss at step  48000 :  5.23191133296
Average loss at step  50000 :  4.98666904318
Nearest to but: and, although, circ, however, or, four, which, two,
Nearest to state: subkey, inventor, feel, refrigerator, rebellious, leche, diodes, independent,
Nearest to five: six, four, three, eight, seven, dasyprocta, agouti, two,
Nearest to their: his, its, the, operatorname, punts, acacia, agouti, biostatistics,
Nearest to as: harbour, circ, agouti, kapoor, six, is, by, aba,
Nearest to other: different, four, many, clock, some, western, three, operatorname,
Nearest to than: or, and, for, rally, sword, simple, but, operatorname,
Nearest to are: were, is, have, be, was, agouti, gad, multivibrator,
Nearest to its: their, the, his, operatorname, dasyprocta, magdeburg, agouti, her,
Nearest to see: kohanim, trinomial, seabirds, aba, originator, erythrocytes, rifles, agouti,
Nearest to and: or, dasyprocta, circ, but, operatorname, six, agouti, four,
Nearest to three: four, six, five, seven, eight, two, one, dasyprocta,
Nearest to it: he, this, she, there, which, amalthea, archie, dasyprocta,
Nearest to war: decide, buprenorphine, ordinarily, northrop, aforementioned, density, slovenes, scenario,
Nearest to however: but, and, almighty, operatorname, technique, educators, that, four,
Nearest to states: defending, agouti, semi, agave, ideology, dasyprocta, hadith, classically,
Average loss at step  52000 :  5.02374834335
Average loss at step  54000 :  5.19878741825
Average loss at step  56000 :  5.04657789743
Average loss at step  58000 :  5.05832611966
Average loss at step  60000 :  4.95238713527
Nearest to but: and, however, or, although, circ, which, operatorname, kapoor,
Nearest to state: subkey, inventor, refrigerator, feel, pulau, ursinus, leche, michelob,
Nearest to five: six, four, three, eight, seven, two, dasyprocta, zero,
Nearest to their: his, its, the, her, operatorname, agouti, punts, a,
Nearest to as: in, circ, agouti, kapoor, by, harbour, is, aba,
Nearest to other: different, many, some, three, operatorname, clock, four, western,
Nearest to than: or, and, for, but, simple, rally, qualification, sword,
Nearest to are: were, is, have, be, gad, agouti, was, fractions,
Nearest to its: their, his, the, her, operatorname, magdeburg, biostatistics, agouti,
Nearest to see: but, kohanim, trinomial, seabirds, erythrocytes, originator, aba, solstice,
Nearest to and: or, circ, operatorname, dasyprocta, agouti, including, but, pulau,
Nearest to three: six, four, five, two, seven, eight, dasyprocta, agouti,
Nearest to it: he, this, she, there, which, they, amalthea, archie,
Nearest to war: buprenorphine, aforementioned, ordinarily, decide, rivaling, northrop, density, ursus,
Nearest to however: but, and, that, almighty, operatorname, technique, when, which,
Nearest to states: agouti, agave, defending, dasyprocta, ideology, semi, hadith, kingdom,
Average loss at step  62000 :  5.00446435535
Average loss at step  64000 :  4.85484569466
Average loss at step  66000 :  4.64125810647
Average loss at step  68000 :  5.0052472105
Average loss at step  70000 :  4.88670423937
Nearest to but: and, however, although, or, callithrix, circ, mico, which,
Nearest to state: subkey, inventor, ursinus, refrigerator, pulau, feel, capuchin, leche,
Nearest to five: four, six, eight, three, seven, zero, nine, two,
Nearest to their: its, his, the, her, some, operatorname, agouti, michelob,
Nearest to as: agouti, circ, kapoor, harbour, aba, operatorname, thaler, in,
Nearest to other: different, many, some, three, profits, operatorname, several, fountain,
Nearest to than: or, but, and, simple, qualification, rally, operatorname, wadsworth,
Nearest to are: were, have, is, be, gad, agouti, almighty, was,
Nearest to its: their, his, the, her, operatorname, agouti, ajanta, biostatistics,
Nearest to see: kohanim, but, trinomial, methodist, entire, originator, clo, aba,
Nearest to and: or, circ, callithrix, but, dasyprocta, operatorname, thaler, agouti,
Nearest to three: six, five, four, seven, eight, two, dasyprocta, one,
Nearest to it: he, she, this, there, which, they, amalthea, abakan,
Nearest to war: callithrix, ordinarily, buprenorphine, aforementioned, rivaling, decide, northrop, kapoor,
Nearest to however: but, and, that, which, almighty, although, when, though,
Nearest to states: thaler, kingdom, agouti, agave, defending, semi, ideology, dasyprocta,
Average loss at step  72000 :  4.7651747334
Average loss at step  74000 :  4.80770084751
Average loss at step  76000 :  4.72523927844
Average loss at step  78000 :  4.80518674481
Average loss at step  80000 :  4.79347246975
Nearest to but: however, and, although, callithrix, mico, circ, or, which,
Nearest to state: subkey, feel, pulau, inventor, capuchin, hadith, refrigerator, ursinus,
Nearest to five: four, six, seven, eight, three, two, nine, zero,
Nearest to their: its, his, the, her, our, agouti, operatorname, michelob,
Nearest to as: agouti, circ, kapoor, microcebus, thaler, callithrix, harbour, astoria,
Nearest to other: many, different, some, three, profits, operatorname, western, clock,
Nearest to than: or, but, and, kosar, simple, qualification, operatorname, wadsworth,
Nearest to are: were, is, have, gad, be, agouti, almighty, was,
Nearest to its: their, his, the, her, biostatistics, operatorname, agouti, dasyprocta,
Nearest to see: but, kohanim, trinomial, originator, aba, clo, milestones, yankees,
Nearest to and: or, callithrix, but, operatorname, circ, dasyprocta, thaler, agouti,
Nearest to three: five, four, six, two, seven, eight, dasyprocta, agouti,
Nearest to it: he, this, there, she, which, they, amalthea, abakan,
Nearest to war: ordinarily, callithrix, buprenorphine, aforementioned, rivaling, northrop, decide, lossy,
Nearest to however: but, that, although, when, though, almighty, operatorname, and,
Nearest to states: kingdom, thaler, agouti, defending, agave, dasyprocta, ideology, semi,
Average loss at step  82000 :  4.76438662171
Average loss at step  84000 :  4.76674228442
Average loss at step  86000 :  4.76899926138
9joml ;;‘.Average loss at step  90000 :  4.7310421375
Nearest to but: and, however, or, which, callithrix, although, while, circ,
Ne


==
ij




arest to state: subkey, pulau, capuchin, ursinus, michelob, hadith, feel, refrigerator,
Nearest to five: four, seven, eight, six, three, zero, nine, two,
Nearest to their: its, his, the, her, our, them, agouti, some,
Nearest to as: agouti, by, in, microcebus, harbour, circ, callithrix, aba,
Nearest to other: different, many, some, including, several, profits, operatorname, three,
Nearest to than: or, but, and, kosar, simple, qualification, archaeopteryx, wadsworth,
Nearest to are: were, have, is, be, include, gad, agouti, almighty,
Nearest to its: their, his, the, her, biostatistics, agouti, dasyprocta, operatorname,
Nearest to see: but, kohanim, originator, trinomial, aba, bragi, yankees, clo,
Nearest to and: or, but, callithrix, operatorname, circ, including, however, dasyprocta,
Nearest to three: four, five, two, six, seven, eight, one, dasyprocta,
Nearest to it: he, this, she, there, which, they, amalthea, abakan,
Nearest to war: callithrix, ordinarily, globemaster, aforementioned, buprenorphine, northrop, rivaling, kapoor,
Nearest to however: but, that, and, although, which, calypso, though, operatorname,
Nearest to states: kingdom, thaler, defending, agave, agouti, ashes, state, dasyprocta,
Average loss at step  92000 :  4.66148071206
Average loss at step  94000 :  4.72550550318
Average loss at step  96000 :  4.68572968149
Average loss at step  98000 :  4.59241366351
Average loss at step  100000 :  4.68939590132
Nearest to but: however, and, although, or, while, which, callithrix, circ,
Nearest to state: subkey, pulau, capuchin, ursinus, michelob, leche, inventor, hadith,
Nearest to five: four, seven, three, six, eight, two, zero, nine,
Nearest to their: its, his, the, her, our, them, some, agouti,
Nearest to as: harbour, agouti, circ, aba, operatorname, kapoor, by, goo,
Nearest to other: different, many, some, including, operatorname, profits, several, these,
Nearest to than: or, and, but, kosar, simple, much, archaeopteryx, qualification,
Nearest to are: were, have, is, be, include, these, gad, agouti,
Nearest to its: their, his, the, her, biostatistics, agouti, dasyprocta, apo,
Nearest to see: kohanim, aba, batted, bragi, but, trinomial, originator, milestones,
Nearest to and: or, but, callithrix, circ, dasyprocta, operatorname, agouti, including,
Nearest to three: five, four, six, two, seven, eight, dasyprocta, agouti,
Nearest to it: he, this, there, she, they, which, abakan, amalthea,
Nearest to war: callithrix, globemaster, ordinarily, aforementioned, northrop, buprenorphine, kapoor, rivaling,
Nearest to however: but, although, that, though, which, and, while, calypso,
Nearest to states: kingdom, thaler, agouti, agave, defending, dasyprocta, state, ideology,
Please install sklearn, matplotlib, and scipy to show embeddings.

Process finished with exit code 0

# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Basic word2vec example."""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import collections
import math
import os
import random
import zipfile

import numpy as np
from six.moves import urllib
from six.moves import xrange  # pylint: disable=redefined-builtin
import tensorflow as tf

# Step 1: Download the data.
url = ‘http://mattmahoney.net/dc/‘


def maybe_download(filename, expected_bytes):
  """Download a file if not present, and make sure it‘s the right size."""
  if not os.path.exists(filename):
    filename, _ = urllib.request.urlretrieve(url + filename, filename)
  statinfo = os.stat(filename)
  if statinfo.st_size == expected_bytes:
    print(‘Found and verified‘, filename)
  else:
    print(statinfo.st_size)
    raise Exception(
        ‘Failed to verify ‘ + filename + ‘. Can you get to it with a browser?‘)
  return filename

filename = maybe_download(‘text8.zip‘, 31344016)


# Read the data into a list of strings.
def read_data(filename):
  """Extract the first file enclosed in a zip file as a list of words."""
  with zipfile.ZipFile(filename) as f:
    data = tf.compat.as_str(f.read(f.namelist()[0])).split()
  return data

vocabulary = read_data(filename)
print(‘Data size‘, len(vocabulary))

# Step 2: Build the dictionary and replace rare words with UNK token.
vocabulary_size = 50000


def build_dataset(words, n_words):
  """Process raw inputs into a dataset."""
  count = [[‘UNK‘, -1]]
  count.extend(collections.Counter(words).most_common(n_words - 1))
  dictionary = dict()
  for word, _ in count:
    dictionary[word] = len(dictionary)
  data = list()
  unk_count = 0
  for word in words:
    if word in dictionary:
      index = dictionary[word]
    else:
      index = 0  # dictionary[‘UNK‘]
      unk_count += 1
    data.append(index)
  count[0][1] = unk_count
  reversed_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
  return data, count, dictionary, reversed_dictionary

data, count, dictionary, reverse_dictionary = build_dataset(vocabulary,
                                                            vocabulary_size)
del vocabulary  # Hint to reduce memory.
print(‘Most common words (+UNK)‘, count[:5])
print(‘Sample data‘, data[:10], [reverse_dictionary[i] for i in data[:10]])

data_index = 0

# Step 3: Function to generate a training batch for the skip-gram model.
def generate_batch(batch_size, num_skips, skip_window):
  global data_index
  assert batch_size % num_skips == 0
  assert num_skips <= 2 * skip_window
  batch = np.ndarray(shape=(batch_size), dtype=np.int32)
  labels = np.ndarray(shape=(batch_size, 1), dtype=np.int32)
  span = 2 * skip_window + 1  # [ skip_window target skip_window ]
  buffer = collections.deque(maxlen=span)
  if data_index + span > len(data):
    data_index = 0
  buffer.extend(data[data_index:data_index + span])
  data_index += span
  for i in range(batch_size // num_skips):
    target = skip_window  # target label at the center of the buffer
    targets_to_avoid = [skip_window]
    for j in range(num_skips):
      while target in targets_to_avoid:
        target = random.randint(0, span - 1)
      targets_to_avoid.append(target)
      batch[i * num_skips + j] = buffer[skip_window]
      labels[i * num_skips + j, 0] = buffer[target]
    if data_index == len(data):
      buffer[:] = data[:span]
      data_index = span
    else:
      buffer.append(data[data_index])
      data_index += 1
  # Backtrack a little bit to avoid skipping words in the end of a batch
  data_index = (data_index + len(data) - span) % len(data)
  return batch, labels

batch, labels = generate_batch(batch_size=8, num_skips=2, skip_window=1)
for i in range(8):
  print(batch[i], reverse_dictionary[batch[i]],
        ‘->‘, labels[i, 0], reverse_dictionary[labels[i, 0]])

# Step 4: Build and train a skip-gram model.

batch_size = 128
embedding_size = 128  # Dimension of the embedding vector.
skip_window = 1       # How many words to consider left and right.
num_skips = 2         # How many times to reuse an input to generate a label.

# We pick a random validation set to sample nearest neighbors. Here we limit the
# validation samples to the words that have a low numeric ID, which by
# construction are also the most frequent.
valid_size = 16     # Random set of words to evaluate similarity on.
valid_window = 100  # Only pick dev samples in the head of the distribution.
valid_examples = np.random.choice(valid_window, valid_size, replace=False)
num_sampled = 64    # Number of negative examples to sample.

graph = tf.Graph()

with graph.as_default():

  # Input data.
  train_inputs = tf.placeholder(tf.int32, shape=[batch_size])
  train_labels = tf.placeholder(tf.int32, shape=[batch_size, 1])
  valid_dataset = tf.constant(valid_examples, dtype=tf.int32)

  # Ops and variables pinned to the CPU because of missing GPU implementation
  with tf.device(‘/cpu:0‘):
    # Look up embeddings for inputs.
    embeddings = tf.Variable(
        tf.random_uniform([vocabulary_size, embedding_size], -1.0, 1.0))
    embed = tf.nn.embedding_lookup(embeddings, train_inputs)

    # Construct the variables for the NCE loss
    nce_weights = tf.Variable(
        tf.truncated_normal([vocabulary_size, embedding_size],
                            stddev=1.0 / math.sqrt(embedding_size)))
    nce_biases = tf.Variable(tf.zeros([vocabulary_size]))

  # Compute the average NCE loss for the batch.
  # tf.nce_loss automatically draws a new sample of the negative labels each
  # time we evaluate the loss.
  loss = tf.reduce_mean(
      tf.nn.nce_loss(weights=nce_weights,
                     biases=nce_biases,
                     labels=train_labels,
                     inputs=embed,
                     num_sampled=num_sampled,
                     num_classes=vocabulary_size))

  # Construct the SGD optimizer using a learning rate of 1.0.
  optimizer = tf.train.GradientDescentOptimizer(1.0).minimize(loss)

  # Compute the cosine similarity between minibatch examples and all embeddings.
  norm = tf.sqrt(tf.reduce_sum(tf.square(embeddings), 1, keep_dims=True))
  normalized_embeddings = embeddings / norm
  valid_embeddings = tf.nn.embedding_lookup(
      normalized_embeddings, valid_dataset)
  similarity = tf.matmul(
      valid_embeddings, normalized_embeddings, transpose_b=True)

  # Add variable initializer.
  init = tf.global_variables_initializer()

# Step 5: Begin training.
num_steps = 100001

with tf.Session(graph=graph) as session:
  # We must initialize all variables before we use them.
  init.run()
  print(‘Initialized‘)

  average_loss = 0
  for step in xrange(num_steps):
    batch_inputs, batch_labels = generate_batch(
        batch_size, num_skips, skip_window)
    feed_dict = {train_inputs: batch_inputs, train_labels: batch_labels}

    # We perform one update step by evaluating the optimizer op (including it
    # in the list of returned values for session.run()
    _, loss_val = session.run([optimizer, loss], feed_dict=feed_dict)
    average_loss += loss_val

    if step % 2000 == 0:
      if step > 0:
        average_loss /= 2000
      # The average loss is an estimate of the loss over the last 2000 batches.
      print(‘Average loss at step ‘, step, ‘: ‘, average_loss)
      average_loss = 0

    # Note that this is expensive (~20% slowdown if computed every 500 steps)
    if step % 10000 == 0:
      sim = similarity.eval()
      for i in xrange(valid_size):
        valid_word = reverse_dictionary[valid_examples[i]]
        top_k = 8  # number of nearest neighbors
        nearest = (-sim[i, :]).argsort()[1:top_k + 1]
        log_str = ‘Nearest to %s:‘ % valid_word
        for k in xrange(top_k):
          close_word = reverse_dictionary[nearest[k]]
          log_str = ‘%s %s,‘ % (log_str, close_word)
        print(log_str)
  final_embeddings = normalized_embeddings.eval()

# Step 6: Visualize the embeddings.


def plot_with_labels(low_dim_embs, labels, filename=‘tsne.png‘):
  assert low_dim_embs.shape[0] >= len(labels), ‘More labels than embeddings‘
  plt.figure(figsize=(18, 18))  # in inches
  for i, label in enumerate(labels):
    x, y = low_dim_embs[i, :]
    plt.scatter(x, y)
    plt.annotate(label,
                 xy=(x, y),
                 xytext=(5, 2),
                 textcoords=‘offset points‘,
                 ha=‘right‘,
                 va=‘bottom‘)

  plt.savefig(filename)

try:
  # pylint: disable=g-import-not-at-top
  from sklearn.manifold import TSNE
  import matplotlib.pyplot as plt

  tsne = TSNE(perplexity=30, n_components=2, init=‘pca‘, n_iter=5000, method=‘exact‘)
  plot_only = 500
  low_dim_embs = tsne.fit_transform(final_embeddings[:plot_only, :])
  labels = [reverse_dictionary[i] for i in xrange(plot_only)]
  plot_with_labels(low_dim_embs, labels)

except ImportError:
  print(‘Please install sklearn, matplotlib, and scipy to show embeddings.‘)