def __init__(self, is_training, config):
self.batch_size = batch_size = config.batch_size # batch_size
self.num_steps = num_steps = config.num_steps #
size = config.hidden_size # 隐藏层
vocab_size = config.vocab_size # 词表size
# 输入占位符
self._input_data = tf.placeholder(tf.int32, [batch_size, num_steps])
self._targets = tf.placeholder(tf.int32, [batch_size, num_steps])
lstm_cell = rnn_cell.BasicLSTMCell(size, forget_bias=0.0)
if is_training and config.keep_prob < 1:
lstm_cell = rnn_cell.DropoutWrapper(
lstm_cell, output_keep_prob=config.keep_prob)
cell = rnn_cell.MultiRNNCell([lstm_cell] * config.num_layers)
self._initial_state = cell.zero_state(batch_size, tf.float32)
with tf.device("/cpu:0"):
embedding = tf.get_variable("embedding", [vocab_size, size])
inputs = tf.nn.embedding_lookup(embedding, self._input_data)
if is_training and config.keep_prob < 1:
inputs = tf.nn.dropout(inputs, config.keep_prob)
outputs = []
states = []
state = self._initial_state
with tf.variable_scope("RNN"):
for time_step in range(num_steps):
if time_step > 0:
tf.get_variable_scope().reuse_variables()
(cell_output, state) = cell(inputs[:, time_step, :], state)
outputs.append(cell_output)
states.append(state)
output = tf.reshape(tf.concat(outputs, 1), [-1, size])
softmax_w = tf.get_variable("softmax_w", [size, vocab_size])
softmax_b = tf.get_variable("softmax_b", [vocab_size])
logits = tf.matmul(output, softmax_w) + softmax_b
loss = tf.contrib.legacy_seq2seq.sequence_loss_by_example(
[logits], [tf.reshape(self._targets, [-1])],
[tf.ones([batch_size * num_steps])], vocab_size)
self._cost = cost = tf.reduce_sum(loss) / batch_size
self._final_state = states[-1]
if not is_training:
return
self._lr = tf.Variable(0.0, trainable=False)
tvars = tf.trainable_variables()
grads, _ = tf.clip_by_global_norm(tf.gradients(cost, tvars),
config.max_grad_norm)
optimizer = tf.train.GradientDescentOptimizer(self.lr)
self._train_op = optimizer.apply_gradients(zip(grads, tvars))
PS_OPS = ['Variable', 'VariableV2', 'AutoReloadVariable']
def assign_to_device(device, ps_device='/cpu:0'):
def _assign(op):
node_def = op if isinstance(op, tf.NodeDef) else op.node_def
if node_def.op in PS_OPS:
return "/" + ps_device
else:
return device
# 所有的Graph的操作默认放在CPU上
with tf.device('/cpu:0'):
tower_grads = []
reuse_vars = False
# tf Graph input
X = tf.placeholder(tf.float32, [None, num_input])
Y = tf.placeholder(tf.float32, [None, num_classes])
# 遍历所有GPU并构建自己的计算图
for i in range(num_gpus):
with tf.device(
assign_to_device('/gpu:{}'.format(i), ps_device='/cpu:0')):
# Split data between GPUs
_x = X[i * batch_size:(i + 1) * batch_size]
_y = Y[i * batch_size:(i + 1) * batch_size]
# 由于dropout在训练和预测阶段是不同的行为，所以需要创建两个独立的图来共享相同参数
logits_train = conv_net(_x,
A = np.random.rand(10000, 10000).astype('float32')
B = np.random.rand(10000, 10000).astype('float32')
# 创建图存储结果
c1 = []
c2 = []
def matpow(M, n):
if n < 1:
return M
else:
return tf.matmul(M, matpow(M, n - 1))
"""单GPU情况"""
with tf.device('/gpu:0'):
a = tf.placeholder(tf.float32, [10000, 10000])
b = tf.placeholder(tf.float32, [10000, 10000])
c1.append(matpow(a, n))
c2.append(matpow(b, n))
with tf.device("/cpu:0"):
sum = tf.add_n(c1)
t1_1 = datetime.datetime.now()
with tf.Session(config=tf.ConfigProto(
log_device_placement=log_device_placement)) as sess:
sess.run(sum, {a: A, b: B})
t2_1 = datetime.datetime.now()
"""多GPU情况"""
with tf.device('/gpu:0'):
labels[i * num_skips + j, 0] = buffer[context_words]
if data_index == len(data):
buffer.extend(data[0:span])
data_index = span
else:
buffer.append(data[data_index])
data_index += 1
data_index = (data_index + len(data) - span) % len(data)
return batch, labels
# Input data
X = tf.placeholder(tf.int32, shape=[None])
Y = tf.placeholder(tf.int32, shape=[None, 1])
with tf.device("/cpu:0"):
embedding = tf.Variable(tf.random_normal([vocab_size, embedding_size]))
X_embed = tf.nn.embedding_lookup(embedding, X)
# 为NCE loss 构造变量
nce_weights = tf.Variable(tf.random_normal([vocab_size, embedding_size]))
nce_biases = tf.Variable(tf.zeros([vocab_size]))
# 为每个batch计算平均的nce loss
loss_op = tf.reduce_mean(
tf.nn.nce_loss(weights=nce_weights,
biases=nce_biases,
labels=Y,
inputs=X_embed,
num_sampled=num_sampled,
num_classes=vocab_size))
def word2vec_basic(log_dir):
# 创建tensorboard的可视化目录
if not os.path.exists(log_dir):
os.makedirs(log_dir)
# 第一步，下载数据
url = 'http://mattmahoney.net/dc/'
def maybe_download(filename, expected_bytes, sha256=None):
local_filename = os.path.join(gettempdir(), filename)
if not os.path.exists(local_filename):
local_filename, _ = urllib.request.urlretrieve(
url + filename, local_filename)
statinfo = os.stat(local_filename)
if sha256 and _hash_file(local_filename) != sha256:
raise Exception('Failed to verify ' + local_filename +
' due to hash '
'mismatch. Can you get to it with a browser?')
if statinfo.st_size == expected_bytes:
print("found and verified", filename)
else:
print(statinfo.st_size)
raise Exception('Failed to verify ' + local_filename +
'. Can you get to it with a browser?')
return local_filename
filename = maybe_download(
'text8.zip',
31344016,
sha256=
'a6640522afe85d1963ad56c05b0ede0a0c000dddc9671758a6cc09b7a38e5232')
# 数据转为List<String>
def read_data(filename):
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))
# 第二步，建词典并且把罕见词替换成UNK
vocabulary_size = 50000
def build_dataset(words, n_words):
count = [['UNK', -1]]
count.extend(collections.Counter(words).most_common(n_words - 1))
dictionary = {word: index for index, (word, _) in enumerate(count)}
data = []
unk_count = 0
for word in words:
index = dictionary.get(word, 0)
if 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: 词表中的所有的词的id
# count： 单词和出现次数的map
# dictionary： 单词-->index 的映射
# reverse_dictionary:index -->单词
data, count, dictionary, reversed_dictionary = build_dataset(
vocabulary, vocabulary_size)
del vocabulary
print('Most common words (+UNK)', count[:5])
print('Sample data', data[:10],
[reversed_dictionary[i] for i in data[:10]])
# 针对skip-gram模型生成batch数据
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)
# skip的范围
span = 2 * skip_window + 1
buffer = collections.deque(maxlen=span)
if data_index + span > len(data):
data_index = 0
buffer.extend(data[data_index:data_index + span]) # 向后取一个窗口内的结果
data_index += span
for i in range(batch_size // num_skips):
context_words = [w for w in range(span) if w != skip_window]
words_to_use = random.sample(context_words, num_skips)
for j, context_words in enumerate(words_to_use):
batch[i * num_skips + j] = buffer[skip_window]
labels[i * num_skips + j, 0] = buffer[context_words]
if data_index == len(data):
buffer.extend(data[0: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 - 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], reversed_dictionary[batch[i]], '->', labels[i, 0],
reversed_dictionary[labels[i, 0]])
# 建立并且训练模型
batch_size = 128
embedding_size = 128 # 词向量维度
skip_window = 1 # 考虑左右几个单词
num_skips = 2 # 复用输入生成标签的次数
num_sampled = 64 # 负样本数量
# 采样一个样本的近邻作为随机验证机，将验证集样本限制为 较低id的单词，是比较高频的构造词汇
# 这三个变量用作显示模型准确率，不影响计算。
valid_size = 16 # 用于评估相似性的随机单词集合
valid_window = 100 #
valid_examples = np.random.choice(valid_window, valid_size, replace=False)
graph = tf.Graph()
with graph.as_default():
# 输入数据
with tf.name_scope('input'):
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)
# 操作op和变量variables 固定在CPU上。
with tf.device('/cpu:0'):
with tf.name_scope('embeddings'):
embeddings = tf.Variable(
tf.random_uniform([vocabulary_size, embedding_size], -1.0,
1.0))
embed = tf.nn.embedding_lookup(embeddings, train_inputs)
# 构造NCE损失的变量
with tf.name_scope('weights'):
nce_weights = tf.Variable(
tf.truncated_normal([vocabulary_size, embedding_size],
stddev=1.0 /
math.sqrt(embedding_size)))
with tf.name_scope('biases'):
nce_biases = tf.Variable(tf.zeros([vocabulary_size]))
# 计算该批次的平均nce损失，当评估损失的时候，自动绘制一个新的负样本。
with tf.name_scope('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))
# 汇总损失
tf.summary.scalar('loss', loss)
# 构造SGD
with tf.name_scope('opytimizer'):
optimizer = tf.train.GradientDescentOptimizer(1.0).minimize(loss)
# 计算小批次样本和所有样本之间的余弦相似度
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)
# merge all summary
merged = tf.summary.merge_all()
init = tf.global_variables_initializer()
saver = tf.train.Saver()
# 开始训练
num_steps = 1000001
with tf.compat.v1.Session(graph=graph) as session:
# 写入摘要
writer = tf.summary.FileWriter(log_dir, session.graph)
init.run()
print('inited..')
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
}
# 定义元变量
run_metadata = tf.RunMetadata()
_, summary, loss_val = session.run([optimizer, merged, loss],
feed_dict=feed_dict,
run_metadata=run_metadata)
average_loss += loss_val
writer.add_summary(summary, step)
if step == (num_steps - 1):
writer.add_run_metadata(run_metadata, 'step%d' % step)
if step % 2000 == 0:
if step > 0:
average_loss /= 2000
# 平均损失是对最近的2000个批次样本的估计。
print('Average loss at step ', step, ': ', average_loss)
average_loss = 0
if step % 10000 == 0:
sim = similarity.eval()
for i in range(valid_size):
valid_word = reversed_dictionary[valid_examples[i]]
top_k = 8
nearest = (-sim[i, :]).argsort()[1:top_k + 1]
log_str = 'Nearest to %s:' % valid_word
print(
log_str, ', '.join([
reversed_dictionary[nearest[k]]
for k in range(top_k)
]))
final_embeddings = normalized_embeddings.eval()
# 写下embedding的相应标签
with open(log_dir + '/metadata.tsv', 'w') as f:
for i in range(vocabulary_size):
f.write(reversed_dictionary[i] + '\n')
# 保存checkpoint
saver.save(session, os.path.join(log_dir, 'model.ckpt'))
# 配置Tensorboard
config = projector.ProjectorConfig()
embedding_conf = config.embeddings.add()
embedding_conf.tensor_name = embeddings.name
embedding_conf.metadata_path = os.path.join(log_dir, 'metadata.tsv')
projector.visualize_embeddings(writer, config)
writer.close()
# Step 6: Visualize the embeddings.
# pylint: disable=missing-docstring
# Function to draw visualization of distance between embeddings.
def plot_with_labels(low_dim_embs, labels, filename):
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 = [reversed_dictionary[i] for i in xrange(plot_only)]
plot_with_labels(low_dim_embs, labels,
os.path.join(gettempdir(), 'tsne.png'))
except ImportError as ex:
print(
'Please install sklearn, matplotlib, and scipy to show embeddings.'
)
print(ex)
def _variable_on_cpu(name, shape, initializer):
# 保存参数到cpu
with tf.device('/cpu:0'):
var = tf.get_variable(name, shape, initializer=initializer)
return var