def save_test_data(y1, y2, y3, i):
sen_y1 = data_processor.getSentence(y1, vocab)[0]
sen_y2 = data_processor.getSentence(y2, vocab)[0]
sen_y3 = data_processor.getSentence(y3, vocab)
data_processor.saveData('\nQuestion ' + str(i + 1) + ':\n' + sen_y1)
data_processor.saveData('\nPositive Answer:\n' + sen_y2)
data_processor.saveData('\nNegative Answers:')
for j in range(4):
data_processor.saveData('\n' + str(j + 1) + ' ' + sen_y3[j])
return
def save_train_data(x1, x2, x3):
sen_x1 = data_processor.getSentence(x1, vocab)
sen_x2 = data_processor.getSentence(x2, vocab)
sen_x3 = data_processor.getSentence(x3, vocab)
for j in range(4):
data_processor.saveData('\nQuestion ' + str(j + 1) + ':\n' +
sen_x1[j])
data_processor.saveData('\nPositive Answer' + ':\n' + sen_x2[j])
data_processor.saveData('\nNegative Answer' + ':\n' + sen_x3[j])
return
def save_data_losses(x1, x2, x3, losses):
sen_x1 = data_processor.getSentence(x1, vocab)
sen_x2 = data_processor.getSentence(x2, vocab)
sen_x3 = data_processor.getSentence(x3, vocab)
# print (np.shape(losses),losses)
num = 0
for k in range(len(losses)):
# print ("losses", np.shape(losses), type(losses))
if (losses[k] != 0.0):
data_processor.saveData('\nQuestion_wrong ' + str(num + 1) +
':\n' + sen_x1[k])
data_processor.saveData('\nPositive Answer' + ':\n' +
sen_x2[k])
data_processor.saveData('\nNegative Answer' + ':\n' +
sen_x3[k])
num += 1
if (num == 4):
return
return
def show(sess, vocab, file_path, sequence_length, ratio):
batch_y1, batch_y2, batch_y3, flag_list = data_processor.loadValData(
vocab, file_path, sequence_length, ratio) # batch_size*seq_len
with tf.device('/cpu:0'), tf.name_scope("embedding"):
embedded_chars1 = tf.nn.embedding_lookup(w_embedding, batch_y1)
embedded_chars2 = tf.nn.embedding_lookup(w_embedding, batch_y2)
embedded_chars3 = tf.nn.embedding_lookup(w_embedding, batch_y3)
embedded_chars1_expanded = tf.expand_dims(
embedded_chars1, -1
) # add a dim at the end of the variable. input of image conv has 4 dims.
embedded_chars2_expanded = tf.expand_dims(embedded_chars2, -1)
embedded_chars3_expanded = tf.expand_dims(embedded_chars3, -1)
pooled_outputs1 = []
pooled_outputs2 = []
pooled_outputs3 = []
for i in range((len(filter_sizes))):
conv = tf.nn.conv2d(embedded_chars1_expanded,
w_conv[i],
strides=[1, 1, 1, 1],
padding="VALID",
name="conv")
h = tf.nn.relu(tf.nn.bias_add(conv, b_conv[i]), name="relu")
# print ("h", h.get_shape())
pooled = tf.nn.max_pool(
h,
ksize=[1, sequence_length - filter_sizes[i] + 1, 1, 1],
strides=[1, 1, 1, 1],
padding='VALID',
name="pool") # shape of pooled is [batch_size,1,1,num_filters]
# print ("pooled_outputs"+ str(i), pooled.get_shape())
pooled_outputs1.append(pooled)
conv = tf.nn.conv2d(embedded_chars2_expanded,
w_conv[i],
strides=[1, 1, 1, 1],
padding="VALID",
name="conv")
h = tf.nn.relu(tf.nn.bias_add(conv, b_conv[i]), name="relu")
pooled = tf.nn.max_pool(
h,
ksize=[1, sequence_length - filter_sizes[i] + 1, 1, 1],
strides=[1, 1, 1, 1],
padding='VALID',
name="pool") # shape of pooled is [batch_size,1,1,num_filters]
pooled_outputs2.append(pooled)
conv = tf.nn.conv2d(embedded_chars3_expanded,
w_conv[i],
strides=[1, 1, 1, 1],
padding="VALID",
name="conv")
# print('\n--- shape of cov is {}'.format(conv.get_shape()))
# Apply nonlinearity
h = tf.nn.relu(tf.nn.bias_add(conv, b_conv[i]), name="relu")
# Max-pooling over the outputs
pooled = tf.nn.max_pool(
h,
ksize=[1, sequence_length - filter_sizes[i] + 1, 1, 1],
strides=[1, 1, 1, 1],
padding='VALID',
name="pool") # shape of pooled is [batch_size,1,1,num_filters]
pooled_outputs3.append(pooled)
# reshape the outputs to combine all the pooled features
num_filters_total = num_filters * len(filter_sizes)
# print ("pooled_outputs1",len(pooled_outputs1), pooled_outputs1[0].get_shape(),pooled_outputs1[1].get_shape(),pooled_outputs1[2].get_shape(),pooled_outputs1[3].get_shape())
h_pooled1 = tf.concat(pooled_outputs1,
3) # the 4th dim corresponds to num_filters
h_pooled1_flat = tf.reshape(h_pooled1, [-1, num_filters_total])
# self.h_pooled1_flat = tf.nn.dropout(self.h_pooled1_reshape,1.0)
# print('\n--- shape of h_pooled1_flat {}'.format(self.h_pooled1_flat.get_shape()))
h_pooled2 = tf.concat(pooled_outputs2, 3)
h_pooled2_flat = tf.reshape(h_pooled2, [-1, num_filters_total])
# self.h_pooled2_flat = tf.nn.dropout(self.h_pooled2_reshape,1.0)
h_pooled3 = tf.concat(pooled_outputs3, 3)
h_pooled3_flat = tf.reshape(h_pooled3, [-1, num_filters_total])
# self.h_pooled3_flat = tf.nn.dropout(self.h_pooled3_reshape,1.0)
len_pooled1 = tf.sqrt(
tf.reduce_sum(tf.multiply(h_pooled1_flat, h_pooled1_flat),
1)) # length of quesiton vectors
# print('\n--- shape of len_pooled1 {}'.format(len_pooled1.get_shape()))
len_pooled2 = tf.sqrt(
tf.reduce_sum(tf.multiply(h_pooled2_flat, h_pooled2_flat),
1)) # length of positive answer vectors
len_pooled3 = tf.sqrt(
tf.reduce_sum(tf.multiply(h_pooled3_flat, h_pooled3_flat),
1)) # length of negative answer vectors
mul_12 = tf.reduce_sum(tf.multiply(h_pooled1_flat, h_pooled2_flat),
1) # wisely multiple vectors, 向量的点积
# print('\n--- shape of mul_12 {}'.format(mul_12.get_shape()))
mul_13 = tf.reduce_sum(tf.multiply(h_pooled1_flat, h_pooled3_flat), 1)
# output
cos_12 = tf.div(
mul_12, tf.multiply(len_pooled1, len_pooled2),
name="scores") # computes the angle between the two vectors
cos_13 = tf.div(mul_13, tf.multiply(len_pooled1, len_pooled3))
max_idx = tf.argmax(cos_13, 0)
pos = sess.run(cos_12)
neg = sess.run(cos_13)
max_idx = sess.run(max_idx)
# tmp1 = neg[max_idx]
if (pos[0] > neg[max_idx]):
print('\033[1;31;34m')
print "Correct prediction"
print "Question:", '\033[0m'
print data_processor.getSentence(batch_y1,
vocab)[0], '\033[1;31;34m'
print "Correct Answer:", '\033[0m'
print data_processor.getSentence(batch_y2,
vocab)[0], '\033[1;31;34m'
print "Predicted Answer:", '\033[0m'
print data_processor.getSentence(batch_y2, vocab)[0], '\033[0m'
else:
print('\033[1;31;34m')
print "False prediction"
print "Question:", '\033[0m'
print data_processor.getSentence(batch_y1,
vocab)[0], '\033[1;31;34m'
print "Correct Answer:", '\033[0m'
print data_processor.getSentence(batch_y2,
vocab)[0], '\033[1;31;34m'
print "Predicted Answer:", '\033[0m'
print data_processor.getSentence(batch_y3,
vocab)[max_idx], '\033[0m'
