[TensorFlow深度學習深入]實戰一·使用embedding_lookup模組對Word2Vec訓練儲存與簡單使用

• Word2Vec簡介

One hot representation用來表示詞向量非常簡單，但是卻有很多問題。最大的問題是我們的詞彙表一般都非常大，比如達到百萬級別，這樣每個詞都用百萬維的向量來表示簡直是記憶體的災難。這樣的向量其實除了一個位置是1，其餘的位置全部都是0，表達的效率不高，能不能把詞向量的維度變小呢？

Dristributed representation可以解決One hot representation的問題，它的思路是通過訓練，將每個詞都對映到一個較短的詞向量上來。所有的這些詞向量就構成了向量空間，進而可以用普通的統計學的方法來研究詞與詞之間的關係。這個較短的詞向量維度是多大呢？這個一般需要我們在訓練時自己來指定。

本博文就是使用TensorFlow的embedding_lookup模組對Word2Vec訓練儲存與簡單使用的探究。
在此基礎之上，我們就可以使用自己訓練的Word2Vec進行RNN處理應用。
此實戰要用到的資料集為text8.zip

• tf.nn.embedding_lookup介紹

  tf.nn.embedding_lookup(params,ids, partition_strategy=’mod’, name=None, validate_indices=True,max_norm=None)

根據ids中的id,尋找params中的對應元素,可以理解為索引,所以ids中元素值不能超出params的第一維的維數值。
比如,ids=[1,3,5],則找出params中下標為1,3,5的向量組成一個矩陣返回。
embedding_lookup不是簡單的查表，id對應的向量是可以訓練的，訓練引數個數應該是 category num*embedding size，也就是說lookup是一種全連線層。

引數說明:
params: 表示完整的embedding張量,或者除了第一維度之外具有相同形狀的P個張量的列表,表示經分割的嵌入張量。
ids: 一個型別為int32或int64的Tensor,包含要在params中查詢的id

• Word2Vec訓練與儲存

程式碼部分：

# encode : utf - 8
# Copyright 2016 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.
# ==============================================================================
import collections
import pickle
import math
import os
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
import random
import zipfile

import numpy as np
import urllib
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

words = read_data(filename)
print('Data size', len(words))

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

def build_dataset(words):
  count = [['UNK', -1]]
  count.extend(collections.Counter(words).most_common(vocabulary_size - 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
  reverse_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
  return data, count, dictionary, reverse_dictionary

data, count, dictionary, reverse_dictionary = build_dataset(words)
del words  # 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)
  for _ in range(span):
    buffer.append(data[data_index])
    data_index = (data_index + 1) % len(data)
  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]
    buffer.append(data[data_index])
    data_index = (data_index + 1) % 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(1.2).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()
  saver = tf.train.Saver()
  print("Initialized")

  average_loss = 0
  for step in range(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 % 1000 == 0:
      if step > 0:
        average_loss /= 1000
      # 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 range(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 range(top_k):
          close_word = reverse_dictionary[nearest[k]]
          log_str = "%s %s," % (log_str, close_word)
        print(log_str)
  
  final_embeddings = normalized_embeddings.eval()
  saver_path = saver.save(session, './2RNN/3_1Word2Vec/MyModel')
  print("saver path: ",saver_path)
  with open('./2RNN/3_1Word2Vec/tf_128_2.pkl', 'wb') as fw:
    pickle.dump({'embeddings': final_embeddings, 'word2id': dictionary, 'id2word': reverse_dictionary}, fw, protocol=4)


# 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:
  from sklearn.manifold import TSNE
  import matplotlib.pyplot as plt

  tsne = TSNE(perplexity=30, n_components=2, init='pca', n_iter=5000)
  plot_only = 200
  low_dim_embs = tsne.fit_transform(final_embeddings[:plot_only,:])
  labels = [reverse_dictionary[i] for i in range(plot_only)]
  plot_with_labels(low_dim_embs, labels)

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

執行結果：

Average loss at step  1000 :  149.03840727233887
Average loss at step  2000 :  86.77497396659851
Average loss at step  3000 :  61.10482195854187
...
Average loss at step  97000 :  4.575266252994537
Average loss at step  98000 :  4.605689331054688
Average loss at step  99000 :  4.6487927632331845
Average loss at step  100000 :  4.653323454380035
Nearest to or: and, agouti, microcebus, ssbn, dasyprocta, than, clodius, mucus,
Nearest to as: agouti, when, microcebus, ssbn, bpp, amalthea, roshan, michelob,
Nearest to i: we, ii, you, t, subcode, they, tabula, g,
Nearest to they: there, he, we, you, it, these, not, who,
Nearest to and: or, but, microcebus, agouti, mucus, dasyprocta, while, michelob,
Nearest to zero: eight, five, seven, four, six, nine, dasyprocta, michelob,
Nearest to states: nations, bandanese, kingdom, absalom, dasyprocta, aediles, applescript, kv,
Nearest to have: had, has, are, were, be, klister, having, agouti,
Nearest to five: four, six, seven, eight, three, two, zero, nine,
Nearest to used: known, agouti, microcebus, iit, abitibi, spoken, dasyprocta, upanija,
Nearest to an: wernicke, riley, binds, oddly, tunings, rearranged, tamarin, apparition,
Nearest to between: with, within, into, from, in,through, jarman, saracens,
Nearest to time: reginae, year, callithrix, iit, albury, upanija, brahma, microcebus,
Nearest to it: he, she, this, there, they, which,amalthea, microcebus,
Nearest to from: into, through, during, in, within, between, dominican, with,
Nearest to six: four, seven, five, eight, nine, three, two, agouti,
saver path:  ./2RNN/3_1Word2Vec/MyModel

結果分析：

訓練10萬步後，loss由149減少到4.6，每個資料都找到了一個較為適合的語料空間位置。
例如：Nearest to five: four, six, seven, eight, three, two, zero, nine。
區分出了，數字詞彙都在靠近的位置。

• 模型的複用

在上個部分我們訓練的過程中，我們也把訓練的結果儲存到了tf_128_2.pkl檔案中，我們這部分要做的就是把儲存的資料給取出來。

程式碼部分

# -*- coding: utf-8 -*-

import tensorflow as tf
import numpy as np
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
import pickle

with open('./2RNN/3_1Word2Vec/tf_128_2.pkl', 'rb') as fr:
    data = pickle.load(fr)
    final_embeddings = data['embeddings']
    word2id = data['word2id']
    id2word = data['id2word']


print("word2id:",type(word2id),len(word2id))
print("word2id one:",list(word2id.items())[0])
print("id2word:",type(id2word),len(id2word))
print("id2word one:",list(id2word.items())[0])
print("final_embeddings:",type(final_embeddings),final_embeddings.shape)
print("final_embeddings one:",final_embeddings[0])

執行結果

word2id: <class 'dict'> 50000
word2id one: ('UNK', 0)
id2word: <class 'dict'> 50000
id2word one: (0, 'UNK')
final_embeddings: <class 'numpy.ndarray'> (50000,128)
final_embeddings one:
[ 0.07824267  0.02380653 -0.04904078 -0.15769418 -0.03343008 -0.00123829
 -0.00840652  0.11035322  0.05255153 -0.01701773 -0.03454393  0.07412812
  0.12529139  0.08700892  0.13564599  0.06016889 -0.02242458  0.01967838
 -0.08621006  0.19164786  0.05878171  0.150539930.15180601  0.11737475
  0.02684335 -0.02697461  0.02076019 -0.074430790.0905515  -0.00580214
 -0.10034874  0.10663538  0.10468851 -0.0018832  -0.03854908 -0.04377652
 -0.07925367 -0.01276041  0.06139784 -0.04612593 -0.0026719  -0.14129621
  0.03356975 -0.08864117  0.03864674  0.06496057 -0.03393148 -0.18256697
  0.1531667   0.01806654 -0.25479555 -0.0102073  -0.01091281 -0.13244723
  0.03231056 -0.04288295  0.00475867 -0.063878960.16555941 -0.1105833
  0.16233324 -0.01569812 -0.03743415  0.118394350.14104177 -0.06637108
 -0.02597998 -0.05089493  0.05379589  0.02132376 -0.0230114   0.16737887
 -0.07722343  0.06376561 -0.06996173  0.07367135 -0.04434428 -0.05931331
  0.13638481 -0.12992401  0.05051441  0.100753180.1285995   0.03757066
 -0.15496145  0.02049168 -0.02400574  0.04723364 -0.05883536  0.20387387
 -0.01346673  0.09482987  0.02737017  0.079759790.02752302  0.1652701
 -0.06379505 -0.01461394 -0.01188034  0.118714   -0.0942675   0.08787307
 -0.06561033  0.04986798  0.18926224  0.111620020.01565995  0.09576936
 -0.02896462  0.03163688  0.08406845  0.07642328 -0.04427774 -0.03355639
 -0.07277506 -0.20906252 -0.00820385 -0.006069670.02557734  0.03273683
  0.04223491  0.04725773 -0.011081   -0.02940390.04183002 -0.00577809
  0.13359077 -0.02493091]

結果分析
word2id與id2word:都是擁有50000元素的字典變數，看用於id與word的相互轉換。
final_embeddings是一個二維資料擁有50000條資料，每個資料為128的向量，就類似於Mnist手寫資料集裡的784個畫素點，這就是詞向量的實質。後面我們就可以用我們自己訓練的詞向量來做語義分析的處理了。