def train_batch():
source_path = 'data/sample2.txt'
source_text = load_data(source_path)
print("source_text:", source_text)
initial_words = get_tokens(source_text)
vocab = Vocab(initial_tokens=initial_words)
vocab.load_pretrained_embeddings(config.embedding_path_air)
sentence_ids = text_to_ids(source_text, vocab.token2id)
sentence_ids = pad_sentence_batch(sentence_ids, vocab.token2id['<blank>'])
# 常量
print("sentence_ids:", sentence_ids)
sentence_place = tf.placeholder(tf.int32, [None, None])
embed_sentences = embed(vocab, sentence_place)
# embed_sentences = tf.nn.l2_normalize(embed_sentences, axis=1)
host = embed_sentences[:3]
guest = embed_sentences[3:]
similarity = tf.matmul(guest, tf.transpose(host))
similarity = tf.identity(similarity, name='similarity')
probabilities = tf.nn.softmax(similarity)
labels = [0, 1, 0, 0, 0, 1, 1]
train_labels_on_seed = get_dummy(labels[:3], labels[3:])
y = neural_net_label_input(3)
the_arg_max = get_argmax(similarity)
pre = get_predict_label(labels[:3], the_arg_max)
acc = get_accuracy(pre, labels[3:])
with tf.Session() as sess:
# Initializing the variables
sess.run(tf.global_variables_initializer())
sess.run(tf.tables_initializer())
train_acc = sess.run(acc, feed_dict={
sentence_place: sentence_ids,
y: train_labels_on_seed
})
prob = sess.run(probabilities, feed_dict={
sentence_place: sentence_ids
})
print('Training Accuracy: {:.6f}'.format(train_acc))
print("prob: ", prob, end='\n\n')
def similarity_matrix():
source_path = 'data/sample.txt'
source_text = load_data(source_path)
print("source_text:", source_text)
vocab = Vocab()
vocab.load_vocab_from_embedding(config.embedding_path_air)
vocab.load_pretrained_embeddings(config.embedding_path_air)
sentence_ids = text_to_ids(source_text, vocab.token2id)
sentence_ids = pad_sentence_batch(sentence_ids, vocab.token2id['<blank>'])
# 常量
print("sentence_ids:", sentence_ids)
sentence_place = tf.placeholder(tf.int32, [None, None])
embed_sentences = embed(vocab, sentence_place)
embed_sentences = tf.nn.l2_normalize(embed_sentences, axis=1)
# enc_embed_input = tf.contrib.layers.embed_sequence(
# sentence_place, source_vocab_size, vocab.embed_dim)
# tf.contrib.layers.embed_sequence() is for what?
# https://stackoverflow.com/questions/40784656/tf-contrib-layers-embed-sequence-is-for-what
host = embed_sentences[:2]
guest = embed_sentences[2:]
# labels = [[1, 0], [1, 0], [0, 1], [0, 1]]
# difference = host[1]-guest[2]
similarity = tf.matmul(guest, tf.transpose(host))
similarity = tf.identity(similarity, name='similarity')
probabilities = tf.nn.softmax(similarity)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
vector = sess.run(embed_sentences,
feed_dict={sentence_place: sentence_ids})
# print("list of vectors: ", vector)
print("type of vector:", type(vector[0]))
print("vector length:", vector[0].shape)
sim = sess.run(similarity, feed_dict={sentence_place: sentence_ids})
print("sim: ", sim)
prob = sess.run(probabilities,
feed_dict={sentence_place: sentence_ids})
print("prob: ", prob)
def similarity_matrix():
source_path = 'data/sample.txt'
source_text = load_data(source_path)
print("source_text:", source_text)
vocab = Vocab()
vocab.load_vocab_from_embedding(config.embedding_path_air)
vocab.load_pretrained_embeddings(config.embedding_path_air)
sentence_ids = text_to_ids(source_text, vocab.token2id)
sentence_ids = pad_sentence_batch(sentence_ids, vocab.token2id['<blank>'])
# 常量
print("sentence_ids:", sentence_ids)
sentence_place = tf.placeholder(tf.int32, [None, None])
embed_sentences = embed(vocab, sentence_place)
embed_sentences = tf.nn.l2_normalize(embed_sentences, axis=1)
host = embed_sentences[:2]
guest = embed_sentences[2:]
similarity = tf.matmul(guest, tf.transpose(host))
similarity = tf.identity(similarity, name='similarity')
probabilities = tf.nn.softmax(similarity)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
vector = sess.run(embed_sentences, feed_dict={
sentence_place: sentence_ids
})
print("type of vector:", type(vector[0]))
print("vector length:", vector[0].shape)
sim = sess.run(similarity, feed_dict={
sentence_place: sentence_ids
})
print("sim: ", sim)
prob = sess.run(probabilities, feed_dict={
sentence_place: sentence_ids
})
print("prob: ", prob)
def train_batch(epochs):
source_path = 'data/sample.txt'
source_text = load_data(source_path)
print("source_text:", source_text)
vocab = Vocab()
vocab.load_vocab_from_embedding(config.embedding_path)
vocab.load_pretrained_embeddings(config.embedding_path)
sentence_ids = text_to_ids(source_text, vocab.token2id)
sentence_ids = pad_sentence_batch(sentence_ids, vocab.token2id['<blank>'])
# 常量
print("sentence_ids:", sentence_ids)
sentence_place = tf.placeholder(tf.int32, [None, None])
embed_sentences = embed(vocab, sentence_place)
embed_sentences = tf.nn.l2_normalize(embed_sentences, axis=1)
host = embed_sentences[:2]
guest = embed_sentences[2:]
similarity = tf.matmul(guest, tf.transpose(host))
similarity = tf.identity(similarity, name='similarity')
probabilities = tf.nn.softmax(similarity)
labels = [[1, 0], [1, 0], [0, 1], [0, 1]]
y = neural_net_label_input(2)
cost = get_probabilities_cost(similarity, y)
optimizer = get_optimizer_single(cost)
accuracy = get_accuracy(similarity, y)
all_params = tf.trainable_variables()
# variable组成的list
print('Checking the Training on a Single Batch...')
with tf.Session() as sess:
# Initializing the variables
sess.run(tf.global_variables_initializer())
param_num = sum([np.prod(sess.run(tf.shape(v))) for v in all_params])
print('There are {} variables in the model'.format(param_num))
# Training cycle
for epoch in range(epochs):
sess.run(optimizer,
feed_dict={
sentence_place: sentence_ids,
y: labels
})
loss = sess.run(cost,
feed_dict={
sentence_place: sentence_ids,
y: labels
})
train_acc = sess.run(accuracy,
feed_dict={
sentence_place: sentence_ids,
y: labels
})
prob = sess.run(probabilities,
feed_dict={sentence_place: sentence_ids})
if epoch % 100 == 0:
print('Epoch {:>2}: '.format(epoch + 1), end='')
print('Loss: {:>10.4f} Training Accuracy: {:.6f}'.format(
loss, train_acc))
print("prob: ", prob, end='\n\n')