window=5,
min_count=5)
model.save(w2v_fp)
wv = model.wv
data.filter_words(set(model.wv.vocab.keys()))
data.cut_train_and_test(balance=True)
## rnn
batch_size = 1024
num_batch_per_epoch = data.get_num_of_batch(batch_size)
hidden_size = 100
num_epoch = 100
lr = 0.002
rnn = RNN("gru", hidden_size, embedding_size, lr=lr)
rnn.build_graph()
sess = tf.Session()
sess.run(tf.global_variables_initializer())
for epoch in range(num_epoch):
for _ in range(num_batch_per_epoch):
batch = data.get_train_batch(batch_size)
batch_X = [
np.array([wv[w] for w in tweet.words]) for tweet in batch
]
batch_y = [int(tweet.label == "bull") for tweet in batch]
_, _ = rnn.train(batch_X, batch_y, sess)
## accuracy check, train, valid, test sets
train_X = [
np.array([wv[w] for w in tweet.words]) for tweet in data.train
w2i = data.word2index()
data.cut_train_and_test(balance=True)
## rnn
batch_size = 128
num_batch_per_epoch = data.get_num_of_batch(batch_size)
hidden_size = 100
num_epoch = 100
lr = 0.002
embedding_size = 100
## embedding
rnn = RNN("gru", hidden_size, embedding_size, lr=lr)
rnn.build_graph(embedding=True, vocab_size=len(w2i), embedding_size=100)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
for epoch in range(num_epoch):
for _ in range(num_batch_per_epoch):
batch = data.get_train_batch(batch_size)
batch_X = [np.array(tweet.word_indexes) for tweet in batch]
batch_y = [int(tweet.label=="bull") for tweet in batch]
_, _= rnn.train(batch_X, batch_y, sess, embedding=True)
## accuracy check, train, valid, test sets
train_X = [np.array(tweet.word_indexes) for tweet in data.train]
train_y = [int(tweet.label == "bull") for tweet in data.train]
train_ent, train_acc = rnn.cal_accuracy(train_X, train_y, sess, embedding=True)
