attMetricAll = []
labelList = [[]]*1000
Q_value = [[]] * 1000
allAccList = []
finalNum = 10
diceAllList=[]
for preNum in range(0,finalNum+1):
tokenizer = tflearn.data_utils.VocabularyProcessor(word_pad_length,
tokenizer_fn=lambda tokens: [token_parse(x) for x in tokens])
label_dict = VocabDict()
reset_default_graph()
model = SelfAttentive()
with tf.Session() as sess:
#load train data
print('load train data')
words, tags = load_csv('../data/trainTCM/TCM_train_%s.csv'%preName, target_columns=[0], columns_to_ignore=None,
target_dict=label_dict,usePreVector=usePreVector)
vocab_list={}
words = string_parser(words, fit=True)
if FLAGS.shuffle == True:
words, tags = shuffle(words, tags)
word_input = tflearn.data_utils.pad_sequences(words, maxlen=word_pad_length)
# build graph
model.build_graph(n=word_pad_length,usePreVector=usePreVector,vectors=word_vecs)
# Downstream Application
with tf.variable_scope('DownstreamApplication'):
global_step = tf.Variable(0, trainable=False, name='global_step')
learn_rate = tf.train.exponential_decay(lr, global_step, FLAGS.decay_step, 0.95, staircase=True)
labels = tf.placeholder('float32', shape=[None, tag_size])
net = tflearn.fully_connected(model.M, 50, activation='relu')
logits = tflearn.fully_connected(net, tag_size, activation=None)
loss = tf.reduce_sum(tf.nn.sigmoid_cross_entropy_with_logits(labels=labels, logits=logits), axis=1)
# preName = 'MM' #11
# preName = 'JP' #124
# preName = 'ZK' #27
# preName = 'AS' #55
# preName = 'HXZT' #1
# preName = 'ZY' #6
# preName = 'TL' #13
allList = []
model = SelfAttentive()
with tf.Session() as sess:
#load train data
print('load train data')
words, tags = load_csv('../data/trainTCM/TCM_train_%s.csv' % preName,
target_columns=[0],
columns_to_ignore=None,
target_dict=label_dict,
usePreVector=usePreVector)
# print('zz',words)
vocab_list = {}
# zsy 判断是否适用预训练的词向量 start
if usePreVector == True:
input_iter = encode_window.create_document_iter(words)
vocab = encode_window.encode_dictionary(input_iter)
vocab_list = vocab.vocabulary_._mapping
word_vecs = encode_window.load_bin_vec(
"../medicalVector/model/medicalCorpus_50d.model", vocab_list)
word_input = encode_window.encode_word(words, vocab, word_pad_length)
else:
words = string_parser(words, fit=True)
if FLAGS.shuffle == True:
loss = tf.reduce_sum(tf.nn.sigmoid_cross_entropy_with_logits(
labels=labels, logits=logits),
axis=1)
loss = tf.reduce_mean(loss)
params = tf.trainable_variables()
optimizer = tf.train.AdamOptimizer(learn_rate)
grad_and_vars = tf.gradients(loss, params)
opt = optimizer.apply_gradients(zip(grad_and_vars, params),
global_step=global_step)
# Start Training
attMetric = []
sess.run(tf.global_variables_initializer())
words, tags = load_csv('../data/trainTCM/TCM_train_%s.csv' %
preName,
target_columns=[0],
columns_to_ignore=None,
target_dict=label_dict)
words = string_parser(words, fit=True)
if FLAGS.shuffle == True:
words, tags = shuffle(words, tags)
word_input = tflearn.data_utils.pad_sequences(
words, maxlen=word_pad_length)
total = len(word_input)
step_print = int((total / batch_size) / 20)
if FLAGS.train == True:
print('start training')
for epoch_num in range(num_epochs):
epoch_loss = 0
step_loss = 0
p_coef = 0.004
p_loss = p_coef * model.P
loss = loss + p_loss
p_loss = tf.reduce_mean(p_loss)
loss = tf.reduce_mean(loss)
params = tf.trainable_variables()
#clipped_gradients = [tf.clip_by_value(x, -0.5, 0.5) for x in gradients]
optimizer = tf.train.AdamOptimizer(learn_rate)
grad_and_vars = tf.gradients(loss, params)
clipped_gradients, _ = tf.clip_by_global_norm(grad_and_vars, 0.5)
opt = optimizer.apply_gradients(zip(clipped_gradients, params), global_step=global_step)
# Start Training
sess.run(tf.global_variables_initializer())
#这里向下都是为了制作句子矩阵
words, tags = load_csv('./data/train.csv', target_columns=[0], target_dict=label_dict, columns_to_ignore=[1])
#words = string_parser(words, fit=True)#将每句话的word转化成对应词汇表的序号
if FLAGS.shuffle == True:
word_input, tags = shuffle(words, tags)
# word2vec作为词向量，需要用word2vec.py预训练好word2vec.txt
'''
w2v_model = gensim.models.KeyedVectors.load_word2vec_format('./data/pw_20_csv/word2vec.txt', binary=False)
embedding_matrix = np.zeros((len(tokenizer.vocabulary_._mapping) + 1, 100))
for word, i in tokenizer.vocabulary_._mapping.items():
try:
embedding_matrix[i] = w2v_model.wv[word]#建立对应单词的向量矩阵
except:
# words not found in embedding index will be all-zeros.
# print(str(i)+': '+word+' not in w2v_model.')
pass
p_coef = 0.004
p_loss = p_coef * model.P
loss = loss + p_loss
p_loss = tf.reduce_mean(p_loss)
loss = tf.reduce_mean(loss)
params = tf.trainable_variables()
#clipped_gradients = [tf.clip_by_value(x, -0.5, 0.5) for x in gradients]
optimizer = tf.train.AdamOptimizer(learn_rate)
grad_and_vars = tf.gradients(loss, params)
clipped_gradients, _ = tf.clip_by_global_norm(grad_and_vars, 0.5)
opt = optimizer.apply_gradients(zip(clipped_gradients, params), global_step=global_step)
# Start Training
sess.run(tf.global_variables_initializer())
words, tags = load_csv('./data/ag_news_csv/train.csv', target_columns=[0], columns_to_ignore=[1], target_dict=label_dict)
words = string_parser(words, fit=True)
if FLAGS.shuffle == True:
words, tags = shuffle(words, tags)
word_input = tflearn.data_utils.pad_sequences(words, maxlen=word_pad_length)
total = len(word_input)
step_print = int((total/batch_size) / 13)
if FLAGS.train == True:
print('start training')
for epoch_num in range(num_epochs):
epoch_loss = 0
step_loss = 0
for i in range(int(total/batch_size)):
batch_input, batch_tags = (word_input[i*batch_size:(i+1)*batch_size], tags[i*batch_size:(i+1)*batch_size])
train_ops = [opt, loss, learn_rate, global_step]
allAccList = []
finalNum = 10
diceAllList = []
for preNum in range(0, 11):
tokenizer = tflearn.data_utils.VocabularyProcessor(
word_pad_length,
tokenizer_fn=lambda tokens: [token_parse(x) for x in tokens])
label_dict = VocabDict()
reset_default_graph()
model = SelfAttentive()
with tf.Session() as sess:
#load train data
print('load train data')
words, tags = load_csv('../data/trainTCM/TCM_train_%s.csv' %
preName,
target_columns=[0],
columns_to_ignore=None,
target_dict=label_dict,
usePreVector=usePreVector)
vocab_list = {}
words = string_parser(words, fit=True)
if FLAGS.shuffle == True:
words, tags = shuffle(words, tags)
word_input = tflearn.data_utils.pad_sequences(
words, maxlen=word_pad_length)
# build graph
model.build_graph(n=word_pad_length,
usePreVector=usePreVector,
vectors=word_vecs)
# Downstream Application
with tf.variable_scope('DownstreamApplication'):
global_step = tf.Variable(0,