def test_sex():
from config import config
from model_sex_chinese import model
f = open('word2idx_chinese.pickle', 'rb')
word2idx = pickle.load(f)
#f = open('career2idx.pickle', 'rb')
#career2idx = pickle.load(f)
V = len(word2idx)
config = config(V)
model = model(config)
batch_size = config.batch_size
test_data = np.load('test_data_c.npy')
test_data_words = np.load('test_data_words_c.npy')
test_ans = np.load('test_ans_c.npy')
blogs, blogs_vec, ans, loss, accuracy, _ = model.build(is_dropout=False)
configs = tf.ConfigProto()
sess = tf.InteractiveSession(config=configs)
saver = tf.train.Saver()
saver.restore(sess, './save_model_sex_c.ckpt')
lens = test_data.shape[0]
T = int(lens / batch_size)
fin_ac = 0.0
for i in range(T):
data_list = test_data[i * batch_size:(i + 1) * batch_size]
data_list_words = test_data_words[i * batch_size:(i + 1) * batch_size]
ans_list = test_ans[i * batch_size:(i + 1) * batch_size, 1]
ac = sess.run(accuracy,
feed_dict={
blogs: data_list,
blogs_vec: data_list_words,
ans: ans_list
})
fin_ac += ac / T
print('acc on sex: %f' % fin_ac)
def sex():
from config import config
from model_sex_chinese import model
f = open('word2idx_chinese.pickle', 'rb')
word2idx = pickle.load(f)
V = len(word2idx)
config = config(V)
config.batch_size = 1
model = model(config)
[dr, blogs, blogs_vec, ans, loss, accuracy,
pans] = model.build(is_dropout=False)
configs = tf.ConfigProto()
configs.gpu_options.allow_growth = True
sess = tf.InteractiveSession(config=configs)
saver = tf.train.Saver()
saver.restore(sess, './saving/save_model_sex_c.ckpt')
return blogs, blogs_vec, tf.arg_max(pans, 1), sess, dr
def train_career():
from config import config
from model_career_chinese import model
f = open('word2idx_chinese.pickle', 'rb')
word2idx = pickle.load(f)
V = len(word2idx)
config = config(V)
model = model(config)
batch_size = config.batch_size
train_data = np.load('train_data_c.npy')
train_data_words = np.load('train_data_words_c.npy')
train_ans = np.load('train_ans_c.npy')
blogs, blogs_vec, ans, loss, accuracy, _ = model.build(is_dropout=True)
var = tf.trainable_variables()
optim = tf.train.AdamOptimizer(0.001).minimize(loss, var_list=var)
saver = tf.train.Saver()
configs = tf.ConfigProto()
sess = tf.InteractiveSession(config=configs)
init = tf.initialize_all_variables()
sess.run(init)
#saver.restore(sess, './save_model_career_c.ckpt')
epoch = 5001
for i in range(epoch):
seed = np.random.randint(0, train_data.shape[0], size=(batch_size))
data_list = np.array([train_data[w] for w in seed])
ans_list = np.array([train_ans[w][0] for w in seed])
data_list_words = np.array([train_data_words[w] for w in seed])
sess.run(optim,
feed_dict={
blogs: data_list,
blogs_vec: data_list_words,
ans: ans_list
})
if i % 10 == 0:
ac, ls = sess.run([accuracy, loss],
feed_dict={
blogs: data_list,
blogs_vec: data_list_words,
ans: ans_list
})
print("i: %d, acc: %f ,loss: %f" % (i, ac, ls))
if i % 5000 == 0:
saver.save(sess, './save_model_career_c.ckpt')
def train_sex():
from config import config
from model_sex_chinese import model
f = open('word2idx_chinese.pickle', 'rb')
word2idx = pickle.load(f)
V = len(word2idx)
config = config(V)
model = model(config)
batch_size = config.batch_size
train_data = np.load('train_data_c.npy')
train_data_words = np.load('train_data_words_c.npy')
train_ans = np.load('train_ans_c.npy')
test_data = np.load('test_data_c.npy')
test_data_words = np.load('test_data_words_c.npy')
test_ans = np.load('test_ans_c.npy')
drop_rate, blogs, blogs_vec, ans, loss, accuracy, _ = model.build(
is_dropout=True)
var = tf.trainable_variables()
optim = tf.train.AdamOptimizer(0.001).minimize(loss, var_list=var)
saver = tf.train.Saver()
configs = tf.ConfigProto()
sess = tf.InteractiveSession(config=configs)
init = tf.initialize_all_variables()
sess.run(init)
epoch = 10001
for i in range(epoch):
seed = np.random.randint(0, train_data.shape[0], size=(batch_size))
data_list = np.array([train_data[w] for w in seed])
ans_list = np.array([train_ans[w, 1] for w in seed])
data_list_words = np.array([train_data_words[w] for w in seed])
sess.run(optim,
feed_dict={
blogs: data_list,
blogs_vec: data_list_words,
ans: ans_list,
drop_rate: 0.5
})
if i % 100 == 0:
ac, ls = sess.run(
[accuracy, loss],
feed_dict={
blogs: data_list,
blogs_vec: data_list_words,
ans: ans_list,
drop_rate: 1.0
})
print("i: %d, acc: %f ,loss: %f" % (i, ac, ls))
if i % 5000 == 0:
saver.save(sess, './save_model_sex_c.ckpt')
if i % 1000 == 0 and i > 0:
lens = test_data.shape[0]
T = int(lens / batch_size)
fin_ac = 0.0
for j in range(T):
data_list = test_data[j * batch_size:(j + 1) * batch_size]
data_list_words = test_data_words[j * batch_size:(j + 1) *
batch_size]
ans_list = test_ans[j * batch_size:(j + 1) * batch_size, 1]
ac = sess.run(accuracy,
feed_dict={
blogs: data_list,
blogs_vec: data_list_words,
ans: ans_list,
drop_rate: 1.0
})
fin_ac += ac / T
print('%d acc on sex: %f' % (i, fin_ac))