def train(self, data, labels):
self.build_model() #整个模型定义过程，
n_batches = int(ceil(data.shape[0] / self.batch_size))
tf.global_variables_initializer().run() #
t_data, t_labels, v_data, v_labels = data_utils.generate_split(
data, labels, self.val_split) #划分数据为训练、验证集
for epoch in range(1, self.n_epochs + 1):
train_cost = 0
for batch in range(1, n_batches + 1):
X, y = data_utils.generate_batch(t_data, t_labels,
self.batch_size) #每次抽取一个batch
#将数据喂入placeholder
f_dict = {
self.inp: X,
self.labels: y,
self.cur_drop_rate: self.dropout_rate
}
_, cost = self.session.run([self.train_op, self.loss],
feed_dict=f_dict) #得到每个batch的损失
train_cost += cost #累计总损失
sys.stdout.write(
'Epoch %d Cost : %f - Batch %d of %d \r' %
(epoch, cost, batch, n_batches))
sys.stdout.flush()
print
self.test(v_data, v_labels)
def train(self, data, labels):
self.build_model()
n_batches = int(ceil(data.shape[0] / self.batch_size))
tf.initialize_all_variables().run()
t_data, t_labels, v_data, v_labels = data_utils.generate_split(
data, labels, self.val_split)
for epoch in range(1, self.n_epochs + 1):
train_cost = 0
for batch in range(1, n_batches + 1):
X, y = data_utils.generate_batch(t_data, t_labels,
self.batch_size)
f_dict = {
self.inp: X,
self.labels: y,
self.cur_drop_rate: self.dropout_rate
}
_, cost = self.session.run([self.train_op, self.loss],
feed_dict=f_dict)
train_cost += cost
sys.stdout.write(
'Epoch %d Cost : %f - Batch %d of %d \r' %
(epoch, cost, batch, n_batches))
sys.stdout.flush()
print
self.test(v_data, v_labels)
def train(self, data, labels):
self.build_model()
if configuration.config['dataset'] == 'CR':
data = data[6792::]
labels = labels[6792::]
train_data = data[0:6791]
train_labels = labels[0:6791]
n_batches = int(ceil(data.shape[0] / self.batch_size))
tf.global_variables_initializer().run()
t_data, t_labels, v_data, v_labels = data_utils.generate_split(data, labels, self.val_split)
if configuration.config['dataset'] == 'CR':
d1, l1, d2, l2 = data_utils.generate_split(train_data, train_labels, self.val_split)
train_data = np.concatenate([d1, d2],0)
train_labels = np.concatenate([l1, l2],0)
t_data = np.concatenate([t_data, train_data],0)
t_labels = np.concatenate([t_labels, train_labels],0)
for epoch in range(1, self.n_epochs + 1):
train_cost = 0
for batch in range(1, n_batches + 1):
X, y = data_utils.generate_batch(t_data, t_labels, self.batch_size)
f_dict = {
self.inp: X,
self.labels: y,
self.cur_drop_rate: self.dropout_rate
}
_, cost = self.session.run([self.train_op, self.loss], feed_dict=f_dict)
train_cost += cost
sys.stdout.write('Epoch %d Cost : %f - Batch %d of %d \r' % (epoch, cost, batch, n_batches))
sys.stdout.flush()
print(self.test(v_data, v_labels))
def test(self, data, labels):
n_batches = int(ceil(data.shape[0] / self.batch_size))
test_cost = 0
preds = []
ys = []
for batch in range(1, n_batches + 1):
X, Y = data_utils.generate_batch(data, labels, self.batch_size)
f_dict = {self.inp: X, self.labels: Y, self.cur_drop_rate: 1.0}
cost, y = self.session.run([self.loss, self.y], feed_dict=f_dict)
test_cost += cost
sys.stdout.write('Cost : %f - Batch %d of %d \r' %
(cost, batch, n_batches))
sys.stdout.flush()
preds.extend(np.argmax(y, 1))
ys.extend(Y)
print
print "Accuracy", np.mean(
np.asarray(np.equal(ys, preds), dtype='float32')) * 100
# Step 1: Download the data (maybe).
filename = maybe_download('text8.zip', 31344016)
# Step 2: Read and build the dictionary and replace rare words with UNK token.
vocabulary = read_data(filename)
print('Data size', len(vocabulary))
vocabulary_size = 50000
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]])
# Step 3: get batch data generator
data_index = 0
sample_batch, sample_labels, data_index = generate_batch(data, data_index, batch_sz=8, n_skips=2, skip_wd=1)
for i in range(8):
print(sample_batch[i], reverse_dictionary[sample_batch[i]], '->', sample_labels[i, 0],
reverse_dictionary[sample_labels[i, 0]])
# Step 4: Build 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.
num_sampled = 64 # Number of negative examples to sample.
# 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. These 3 variables are used only for
# displaying model accuracy, they don't affect calculation.
valid_size = 16 # Random set of words to evaluate similarity on.