def predict_txt(self):
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
all_num, correct_num = 0, 0
for line in open('./test_with_c1_03.txt', 'r',
encoding='utf-8').readlines():
lines = line.replace('\n', '').split('\t\t')
sent, label, label_word = lines[3::]
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent),
'4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
if label_word == '异常':
all_num += 1
t1, t1_len, poss, t1_re, t1_re_len, t1_char, t1_char_len, t1_rm_char, t1_rm_char_len = pdd.predict(
sent)
result = self.predict(t1, t1_len, poss, t1_re, t1_re_len,
t1_char, t1_char_len, t1_rm_char,
t1_rm_char_len, key_emb)
print(id2label3[result[2][1][0]])
if id2label3[result[2][1][0]] == '失败':
correct_num += 1
print(sent, label_word, "失败")
else:
print(sent, label_word, result)
print(correct_num, all_num)
def predict_sent(self):
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
id2label_list = [id2label1, id2label2, id2label3]
pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
while True:
ss = []
sent = input("输入:")
print(jieba.lcut(sent))
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent),
'4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
t1, t1_len, poss, t1_re, t1_re_len, t1_char, t1_char_len, t1_rm_char, t1_rm_char_len = pdd.predict(
sent)
result = self.predict(t1, t1_len, poss, t1_re, t1_re_len, t1_char,
t1_char_len, t1_rm_char, t1_rm_char_len,
key_emb)
for i, res in enumerate(result):
id2label = id2label_list[i]
# print(res[0][0])
# print(id2label[res[1][0]])
ss.append("_".join([str(res[0][0]), str(id2label[res[1][0]])]))
print(ss)
def predict_cls_semantic(self):
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
id2label_list = [id2label1, id2label2, id2label3]
pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
while True:
ss = []
sent1 = input("输入sent1:")
sent2 = input("输入sent2:")
print(jieba.lcut(sent1))
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent1),
'4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
t1_sent1, t1_len_sent1, poss_sent1, t1_re_sent1, t1_re_len_sent1, t1_char_sent1, t1_char_len_sent1, t1_rm_char_sent1, t1_rm_char_len_sent1 \
= pdd.predict(sent1)
t1_sent2, t1_len_sent2, poss_sent2, t1_re_sent2, t1_re_len_sent2, t1_char_sent2, t1_char_len_sent2, t1_rm_char_sent2, t1_rm_char_len_sent2 \
= pdd.predict(sent2)
feed_dict = {
self.sent_token_neg: t1_sent2,
self.sent_len_neg: t1_len_sent2,
self.sent_char_neg: t1_char_sent2,
self.sent_char_len_neg: t1_char_len_sent2,
self.sent_token: t1_sent1,
self.sent_char: t1_char_sent1,
self.sent_len_char: t1_char_len_sent1,
self.sent_len: t1_len_sent1,
self.dropout: 0.0
}
# print(sent1_word_)
s1, s2, s3, pre1, pre2, pre3 = self.sess.run([
self.smentic_out_1, self.smentic_out_2, self.smentic_out_3,
self.soft_out_1, self.soft_out_2, self.soft_out_3
],
feed_dict=feed_dict)
print(s1, s2, s3)
res = []
for ele in [pre1, pre2, pre3]:
score = np.max(ele, 1)
pre_leb = np.argmax(ele, 1)
res.append([score, pre_leb])
for i, res in enumerate(res):
id2label = id2label_list[i]
# print(res[0][0])
# print(id2label[res[1][0]])
ss.append("_".join([str(res[0][0]), str(id2label[res[1][0]])]))
print(ss)
def predict(args, model_name, restore_path):
intent_dict = pkl.load(open(PATH + "/intent_dict.p", 'rb'))
args.id2intent = {v: k for k, v in intent_dict.items()}
vocab = WordVocab.load_vocab(PATH + args.vocab_path)
poss_vocab = pkl.load(open(PATH + "/poss_vocab.p", 'rb'))
args.num_layers = 1
args.vocab_size = len(vocab)
args.class_nums = len(intent_dict)
args.poss_num = len(poss_vocab)
if args.use_pre_train_emb:
vocab_emb = pkl.load(open('%s_vocab_emb.p' % args.task_name, 'rb'))
args.vocab_emb = vocab_emb
if model_name == 'BaseLSTM':
model = BaseLstm(args, 'BaseLstm')
elif model_name == 'BaseLstmStruct':
model = BaseLstmStruct(args, 'BaseLstmStruct')
elif model_name == 'BaseTransformerStruct':
model = BaseTransformerStruct(args, 'BaseTransformerStruct')
elif model_name == 'cnn':
model = Cnn(args, 'cnn')
elif model_name == 'TransformerCNN':
model = TransformerCNN(args, 'TransformerCNN')
elif model_name == 'LEAM':
model = LEAM(args, 'LEAM')
args.model_name = model_name
model.build_placeholder()
model.build_model()
config = tf.ConfigProto(allow_soft_placement=True)
with tf.Session(config=config) as sess:
sess = model.restore(sess, restore_path)
pdd = PredictDataDeal(vocab=vocab,
seq_len=args.seq_len,
poss_vocab=poss_vocab,
vocab_char=None)
while True:
sent = input("输入:")
t1, t1_len, poss = pdd.predict(sent)
pre_prob, pre_label = model.predict(sess, t1, t1_len, poss)
print(args.id2intent[pre_label[0]], np.max(pre_prob, -1))
def __init__(self, model_name='/'):
self.sess_dict = {}
self.queryObj_dict = {}
config = tf.ConfigProto(allow_soft_placement=True)
model_name_meta = '%s.meta' % model_name
saver = tf.train.import_meta_graph(model_name_meta) # 加载图结构
gragh = tf.get_default_graph() # 获取当前图，为了后续训练时恢复变量
tensor_name_list = [
tensor.name for tensor in gragh.as_graph_def().node
] # 得到当前图中所有变量的名称
for ele in tensor_name_list:
if str(ele).__contains__('out_softmax'):
print(ele)
args_dict = json.load(open('./Configs/BaseLstm.config', 'r'))
self.args = dict_to_object(dict(args_dict))
self.label_vocab1 = pkl.load(open("./label_vocab1.p", 'rb'))
self.label_vocab2 = pkl.load(open("./label_vocab2.p", 'rb'))
self.label_vocab3 = pkl.load(open("./label_vocab3.p", 'rb'))
self.vocab = WordVocab.load_vocab(PATH + self.args.vocab_path)
self.poss_vocab = pkl.load(open("./poss_vocab.p", 'rb'))
with tf.device('/device:GPU:%s' % 0):
self.sent_token = gragh.get_tensor_by_name('sent1_token:0')
self.sent_char = gragh.get_tensor_by_name('sent1_char:0')
self.sent_word_re = gragh.get_tensor_by_name('sent_word_re:0')
self.sent_word_re_char = gragh.get_tensor_by_name(
'sent_word_re_char:0')
self.sent_len = gragh.get_tensor_by_name('sent1_len:0')
self.sent_len_char = gragh.get_tensor_by_name('sent_len_char:0')
self.sent_len_re = gragh.get_tensor_by_name('sent1_len_re:0')
self.sent_len_re_char = gragh.get_tensor_by_name(
'sent1_len_re_char:0')
self.sent_token_neg = gragh.get_tensor_by_name('sent1_token_neg:0')
self.sent_len_neg = gragh.get_tensor_by_name('sent1_len_neg:0')
self.sent_char_neg = gragh.get_tensor_by_name('sent_char_neg:0')
self.sent_char_len_neg = gragh.get_tensor_by_name(
'sent_char_len_neg:0')
self.key_emb = gragh.get_tensor_by_name('key_emb:0')
self.dropout = gragh.get_tensor_by_name('dropout:0')
name = model_name.split('/')[-1].split('_')[0].replace(
"BaseLSTM", "BaseLstm")
try:
self.soft_out_1 = gragh.get_tensor_by_name(
'%s_enc_0/_0/softmax/Softmax:0' % name)
self.soft_out_2 = gragh.get_tensor_by_name(
'%s_enc_1/_1/softmax/Softmax:0' % name)
self.soft_out_3 = gragh.get_tensor_by_name(
'%s_enc_2/_2/softmax/Softmax:0' % name)
except:
self.soft_out_1 = gragh.get_tensor_by_name(
'%s_enc_0/_0/out_softmax/softmax/Softmax:0' % name)
self.soft_out_2 = gragh.get_tensor_by_name(
'%s_enc_1/_1/out_softmax/softmax/Softmax:0' % name)
self.soft_out_3 = gragh.get_tensor_by_name(
'%s_enc_2/_2/out_softmax/softmax/Softmax:0' % name)
try:
self.smentic_out_1 = gragh.get_tensor_by_name(
'%s_enc_0/_0/semantic_out/concat:0' % name)
self.smentic_out_2 = gragh.get_tensor_by_name(
'%s_enc_1/_1/semantic_out/concat:0' % name)
self.smentic_out_3 = gragh.get_tensor_by_name(
'%s_enc_2/_2/semantic_out/concat:0' % name)
except:
pass
self.sess = tf.Session(config=config)
saver.restore(self.sess, '%s' % model_name)
self.pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
class IntentCls(object):
def __init__(self, model_name='/'):
self.sess_dict = {}
self.queryObj_dict = {}
config = tf.ConfigProto(allow_soft_placement=True)
model_name_meta = '%s.meta' % model_name
saver = tf.train.import_meta_graph(model_name_meta) # 加载图结构
gragh = tf.get_default_graph() # 获取当前图，为了后续训练时恢复变量
tensor_name_list = [
tensor.name for tensor in gragh.as_graph_def().node
] # 得到当前图中所有变量的名称
for ele in tensor_name_list:
if str(ele).__contains__('out_softmax'):
print(ele)
args_dict = json.load(open('./Configs/BaseLstm.config', 'r'))
self.args = dict_to_object(dict(args_dict))
self.label_vocab1 = pkl.load(open("./label_vocab1.p", 'rb'))
self.label_vocab2 = pkl.load(open("./label_vocab2.p", 'rb'))
self.label_vocab3 = pkl.load(open("./label_vocab3.p", 'rb'))
self.vocab = WordVocab.load_vocab(PATH + self.args.vocab_path)
self.poss_vocab = pkl.load(open("./poss_vocab.p", 'rb'))
with tf.device('/device:GPU:%s' % 0):
self.sent_token = gragh.get_tensor_by_name('sent1_token:0')
self.sent_char = gragh.get_tensor_by_name('sent1_char:0')
self.sent_word_re = gragh.get_tensor_by_name('sent_word_re:0')
self.sent_word_re_char = gragh.get_tensor_by_name(
'sent_word_re_char:0')
self.sent_len = gragh.get_tensor_by_name('sent1_len:0')
self.sent_len_char = gragh.get_tensor_by_name('sent_len_char:0')
self.sent_len_re = gragh.get_tensor_by_name('sent1_len_re:0')
self.sent_len_re_char = gragh.get_tensor_by_name(
'sent1_len_re_char:0')
self.sent_token_neg = gragh.get_tensor_by_name('sent1_token_neg:0')
self.sent_len_neg = gragh.get_tensor_by_name('sent1_len_neg:0')
self.sent_char_neg = gragh.get_tensor_by_name('sent_char_neg:0')
self.sent_char_len_neg = gragh.get_tensor_by_name(
'sent_char_len_neg:0')
self.key_emb = gragh.get_tensor_by_name('key_emb:0')
self.dropout = gragh.get_tensor_by_name('dropout:0')
name = model_name.split('/')[-1].split('_')[0].replace(
"BaseLSTM", "BaseLstm")
try:
self.soft_out_1 = gragh.get_tensor_by_name(
'%s_enc_0/_0/softmax/Softmax:0' % name)
self.soft_out_2 = gragh.get_tensor_by_name(
'%s_enc_1/_1/softmax/Softmax:0' % name)
self.soft_out_3 = gragh.get_tensor_by_name(
'%s_enc_2/_2/softmax/Softmax:0' % name)
except:
self.soft_out_1 = gragh.get_tensor_by_name(
'%s_enc_0/_0/out_softmax/softmax/Softmax:0' % name)
self.soft_out_2 = gragh.get_tensor_by_name(
'%s_enc_1/_1/out_softmax/softmax/Softmax:0' % name)
self.soft_out_3 = gragh.get_tensor_by_name(
'%s_enc_2/_2/out_softmax/softmax/Softmax:0' % name)
try:
self.smentic_out_1 = gragh.get_tensor_by_name(
'%s_enc_0/_0/semantic_out/concat:0' % name)
self.smentic_out_2 = gragh.get_tensor_by_name(
'%s_enc_1/_1/semantic_out/concat:0' % name)
self.smentic_out_3 = gragh.get_tensor_by_name(
'%s_enc_2/_2/semantic_out/concat:0' % name)
except:
pass
self.sess = tf.Session(config=config)
saver.restore(self.sess, '%s' % model_name)
self.pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
def get_key_info(self, key_dict, sents, intent):
if intent in key_dict:
keys = key_dict[intent]
key_emb = []
for word in sents:
if word in keys:
key_emb.append(5)
else:
key_emb.append(1)
key_emb = [1] + key_emb + [1]
key_emb.extend([0] * self.args.seq_len)
else:
key_emb = [5] * self.args.seq_len
return key_emb[:self.args.seq_len]
def predict(self, t1, t1_len, poss, t1_re, t1_re_len, t1_char, t1_char_len,
t1_rm_char, t1_rm_char_len, key_emb):
feed_dict = {
self.sent_token: t1,
self.sent_len: t1_len,
self.sent_word_re: t1_re,
self.sent_len_re: t1_re_len,
self.sent_char: t1_char,
self.sent_len_char: t1_char_len,
self.sent_word_re_char: t1_rm_char,
self.sent_len_re_char: t1_rm_char_len,
self.key_emb: key_emb,
self.dropout: 0.0
}
# print(sent1_word_)
pre1, pre2, pre3 = self.sess.run(
[self.soft_out_1, self.soft_out_2, self.soft_out_3],
feed_dict=feed_dict)
res = []
for ele in [pre1, pre2, pre3]:
score = np.max(ele, 1)
pre_leb = np.argmax(ele, 1)
res.append([score, pre_leb])
return res
def predict_sent(self):
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
id2label_list = [id2label1, id2label2, id2label3]
pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
while True:
ss = []
sent = input("输入:")
print(jieba.lcut(sent))
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent),
'4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
t1, t1_len, poss, t1_re, t1_re_len, t1_char, t1_char_len, t1_rm_char, t1_rm_char_len = pdd.predict(
sent)
result = self.predict(t1, t1_len, poss, t1_re, t1_re_len, t1_char,
t1_char_len, t1_rm_char, t1_rm_char_len,
key_emb)
for i, res in enumerate(result):
id2label = id2label_list[i]
# print(res[0][0])
# print(id2label[res[1][0]])
ss.append("_".join([str(res[0][0]), str(id2label[res[1][0]])]))
print(ss)
def predict_api(self, sent):
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
id2label_list = [id2label1, id2label2, id2label3]
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent), '4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
t1, t1_len, poss, t1_re, t1_re_len, t1_char, t1_char_len, t1_rm_char, t1_rm_char_len = self.pdd.predict(
sent)
result = self.predict(t1, t1_len, poss, t1_re, t1_re_len, t1_char,
t1_char_len, t1_rm_char, t1_rm_char_len, key_emb)
result_finall = []
dd_dict = {}
for i, res in enumerate(result):
id2label = id2label_list[i]
dd_dict['label_%s' % i] = [id2label[res[1][0]], str(res[0][0])]
# result_finall.append([id2label[res[1][0]]res[0][0]])
return dd_dict
def predict_txt(self):
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
all_num, correct_num = 0, 0
for line in open('./test_with_c1_03.txt', 'r',
encoding='utf-8').readlines():
lines = line.replace('\n', '').split('\t\t')
sent, label, label_word = lines[3::]
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent),
'4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
if label_word == '异常':
all_num += 1
t1, t1_len, poss, t1_re, t1_re_len, t1_char, t1_char_len, t1_rm_char, t1_rm_char_len = pdd.predict(
sent)
result = self.predict(t1, t1_len, poss, t1_re, t1_re_len,
t1_char, t1_char_len, t1_rm_char,
t1_rm_char_len, key_emb)
print(id2label3[result[2][1][0]])
if id2label3[result[2][1][0]] == '失败':
correct_num += 1
print(sent, label_word, "失败")
else:
print(sent, label_word, result)
print(correct_num, all_num)
def predict_cls_semantic(self):
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
id2label_list = [id2label1, id2label2, id2label3]
pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
while True:
ss = []
sent1 = input("输入sent1:")
sent2 = input("输入sent2:")
print(jieba.lcut(sent1))
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent1),
'4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
t1_sent1, t1_len_sent1, poss_sent1, t1_re_sent1, t1_re_len_sent1, t1_char_sent1, t1_char_len_sent1, t1_rm_char_sent1, t1_rm_char_len_sent1 \
= pdd.predict(sent1)
t1_sent2, t1_len_sent2, poss_sent2, t1_re_sent2, t1_re_len_sent2, t1_char_sent2, t1_char_len_sent2, t1_rm_char_sent2, t1_rm_char_len_sent2 \
= pdd.predict(sent2)
feed_dict = {
self.sent_token_neg: t1_sent2,
self.sent_len_neg: t1_len_sent2,
self.sent_char_neg: t1_char_sent2,
self.sent_char_len_neg: t1_char_len_sent2,
self.sent_token: t1_sent1,
self.sent_char: t1_char_sent1,
self.sent_len_char: t1_char_len_sent1,
self.sent_len: t1_len_sent1,
self.dropout: 0.0
}
# print(sent1_word_)
s1, s2, s3, pre1, pre2, pre3 = self.sess.run([
self.smentic_out_1, self.smentic_out_2, self.smentic_out_3,
self.soft_out_1, self.soft_out_2, self.soft_out_3
],
feed_dict=feed_dict)
print(s1, s2, s3)
res = []
for ele in [pre1, pre2, pre3]:
score = np.max(ele, 1)
pre_leb = np.argmax(ele, 1)
res.append([score, pre_leb])
for i, res in enumerate(res):
id2label = id2label_list[i]
# print(res[0][0])
# print(id2label[res[1][0]])
ss.append("_".join([str(res[0][0]), str(id2label[res[1][0]])]))
print(ss)
def predict_xlsx(self):
dd_2 = {}
for ele in open('./test_with_c1_02.txt', 'r',
encoding='utf-8').readlines():
eles = ele.replace('\n', '').split('\t\t')
sent = eles[3]
score = eles[2]
label = eles[1]
if label.__contains__('inter') or label.__contains__('自动续费'):
label = "None"
dd_2[sent] = [label, score]
dd_3 = {}
for ele in open('./test_03.txt', 'r', encoding='utf-8').readlines():
eles = ele.replace('\n', '').split('\t\t')
sent = eles[3]
score = eles[2]
label = eles[1]
if label.__contains__('inter') or label.__contains__('自动续费'):
label = "None"
dd_3[sent] = [label, score]
dd_2 = {}
dd_3 = {}
# config = tf.ConfigProto(allow_soft_placement=True)
path = './cp_4000142_0726.xlsx'
# path = './红包退回规则.xlsx'
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
id2label_list = [id2label1, id2label2, id2label3]
true_label_list = []
pre_label_list = []
fw = open("cp_4000142_0726_自动续费.txt", 'w', encoding='utf-8')
fw_false = open("cp_4000142_0726_自动续费_false.txt",
'w',
encoding='utf-8')
count_dict = {}
# with tf.Session(config=config) as sess:
# sess = model.restore(sess, restore_path)
pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
work_sheet = xlrd.open_workbook(path)
sheet = work_sheet.sheet_by_index(1)
stand_list = [
"自动续费_撤销_None", "自动续费_None_None", "自动续费_退款_None", "自动续费_开通_None",
"自动续费_查询_None", "自动续费_退回规则_None"
]
# stand_list = ["红包_退款_None", "红包_退回规则_None"," 红包_退款_未入账"]
stand_list = [str(e).lower() for e in stand_list]
for i in range(1, sheet.nrows):
try:
sent = sheet.cell_value(i, 2)
label = sheet.cell_value(i, 3)
except:
sent = sheet.cell_value(i, 0)
label = sheet.cell_value(i, 1)
# entity = entity_detect(sent)
flag = True
t1, t1_len, poss, t1_re, t1_re_len, t1_char, t1_char_len, t1_rm_char, t1_rm_char_len = pdd.predict(
sent)
# result=model.predict(sess,t1,t1_len,poss,t1_re,t1_re_len,t1_char,t1_char_len,t1_rm_char,t1_rm_char_len)
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent),
'4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
result = self.predict(t1, t1_len, poss, t1_re, t1_re_len, t1_char,
t1_char_len, t1_rm_char, t1_rm_char_len,
key_emb)
ss = []
pp_ss = []
for j, res in enumerate(result):
id2label = id2label_list[j]
# print(res[0][0])
# print(id2label[res[1][0]])
if j == 1:
if sent in dd_2:
ss.append(str(dd_2[sent][1]) + '_' + dd_2[sent][0])
pp_ss.append(dd_2[sent][0])
else:
if res[0][0] <= 0.8:
ss.append("0.0" + '_' + 'None')
pp_ss.append('None')
else:
ss.append(
str(res[0][0]) + '_' + id2label[res[1][0]])
pp_ss.append(id2label[res[1][0]])
if j == 2:
if sent in dd_3:
ss.append(str(dd_3[sent][1]) + '_' + dd_3[sent][0])
pp_ss.append(dd_3[sent][0])
else:
if res[0][0] <= 0.5:
ss.append("0.0" + '_' + 'None')
pp_ss.append('None')
else:
ss.append(
str(res[0][0]) + '_' + id2label[res[1][0]])
pp_ss.append(id2label[res[1][0]])
if j == 0:
if res[0][0] <= 0.5:
ss.append("0.0" + '_' + 'None')
pp_ss.append('None')
else:
ss.append(str(res[0][0]) + '_' + id2label[res[1][0]])
pp_ss.append(id2label[res[1][0]])
pp_str = '_'.join(pp_ss).lower()
print(sent, '\t\t', pp_str)
if pp_str in stand_list:
fw.write("True" + '\t\t' + sent + '\t\t' + str(label) + '\n')
if int(label) == 3:
fw_false.write("True" + '\t\t' + sent + '\t\t' +
str(label) + '\n')
fw_false.write(" ".join(ss) + '\n')
fw_false.write('\n')
else:
if int(label) in [1, 2, 5]:
fw.write("False" + '\t\t' + sent + '\t\t' + str(label) +
'\n')
fw_false.write("False" + '\t\t' + sent + '\t\t' +
str(label) + '\n')
fw_false.write(" ".join(ss) + '\n')
fw_false.write('\n')
fw.write(" ".join(ss) + '\n')
fw.write('\n')
if int(label) == 3:
if pp_str in stand_list:
count_dict['3_correct'] = count_dict.get('3_correct',
0) + 1
else:
count_dict['3_error'] = count_dict.get('3_error', 0) + 1
if int(label) == 1:
if pp_str in stand_list:
count_dict['1_correct'] = count_dict.get('1_correct',
0) + 1
else:
count_dict['1_error'] = count_dict.get('1_error', 0) + 1
if int(label) == 2:
if pp_str in stand_list:
count_dict['2_correct'] = count_dict.get('2_correct',
0) + 1
else:
count_dict['2_error'] = count_dict.get('2_error', 0) + 1
if int(label) == 5:
if pp_str in stand_list:
count_dict['5_correct'] = count_dict.get('5_correct',
0) + 1
else:
count_dict['5_error'] = count_dict.get('5_error', 0) + 1
print(count_dict)
def predict_xlsx(self):
dd_2 = {}
for ele in open('./test_with_c1_02.txt', 'r',
encoding='utf-8').readlines():
eles = ele.replace('\n', '').split('\t\t')
sent = eles[3]
score = eles[2]
label = eles[1]
if label.__contains__('inter') or label.__contains__('自动续费'):
label = "None"
dd_2[sent] = [label, score]
dd_3 = {}
for ele in open('./test_03.txt', 'r', encoding='utf-8').readlines():
eles = ele.replace('\n', '').split('\t\t')
sent = eles[3]
score = eles[2]
label = eles[1]
if label.__contains__('inter') or label.__contains__('自动续费'):
label = "None"
dd_3[sent] = [label, score]
dd_2 = {}
dd_3 = {}
# config = tf.ConfigProto(allow_soft_placement=True)
path = './cp_4000142_0726.xlsx'
# path = './红包退回规则.xlsx'
id2label1 = {v: k for k, v in self.label_vocab1.items()}
id2label2 = {v: k for k, v in self.label_vocab2.items()}
id2label3 = {v: k for k, v in self.label_vocab3.items()}
id2label_list = [id2label1, id2label2, id2label3]
true_label_list = []
pre_label_list = []
fw = open("cp_4000142_0726_自动续费.txt", 'w', encoding='utf-8')
fw_false = open("cp_4000142_0726_自动续费_false.txt",
'w',
encoding='utf-8')
count_dict = {}
# with tf.Session(config=config) as sess:
# sess = model.restore(sess, restore_path)
pdd = PredictDataDeal(vocab=self.vocab,
seq_len=self.args.seq_len,
poss_vocab=self.poss_vocab)
work_sheet = xlrd.open_workbook(path)
sheet = work_sheet.sheet_by_index(1)
stand_list = [
"自动续费_撤销_None", "自动续费_None_None", "自动续费_退款_None", "自动续费_开通_None",
"自动续费_查询_None", "自动续费_退回规则_None"
]
# stand_list = ["红包_退款_None", "红包_退回规则_None"," 红包_退款_未入账"]
stand_list = [str(e).lower() for e in stand_list]
for i in range(1, sheet.nrows):
try:
sent = sheet.cell_value(i, 2)
label = sheet.cell_value(i, 3)
except:
sent = sheet.cell_value(i, 0)
label = sheet.cell_value(i, 1)
# entity = entity_detect(sent)
flag = True
t1, t1_len, poss, t1_re, t1_re_len, t1_char, t1_char_len, t1_rm_char, t1_rm_char_len = pdd.predict(
sent)
# result=model.predict(sess,t1,t1_len,poss,t1_re,t1_re_len,t1_char,t1_char_len,t1_rm_char,t1_rm_char_len)
key_emb = self.get_key_info(KEY_WORD_DICT, jieba.lcut(sent),
'4000142')
key_emb = np.expand_dims(np.array(key_emb), 0)
result = self.predict(t1, t1_len, poss, t1_re, t1_re_len, t1_char,
t1_char_len, t1_rm_char, t1_rm_char_len,
key_emb)
ss = []
pp_ss = []
for j, res in enumerate(result):
id2label = id2label_list[j]
# print(res[0][0])
# print(id2label[res[1][0]])
if j == 1:
if sent in dd_2:
ss.append(str(dd_2[sent][1]) + '_' + dd_2[sent][0])
pp_ss.append(dd_2[sent][0])
else:
if res[0][0] <= 0.8:
ss.append("0.0" + '_' + 'None')
pp_ss.append('None')
else:
ss.append(
str(res[0][0]) + '_' + id2label[res[1][0]])
pp_ss.append(id2label[res[1][0]])
if j == 2:
if sent in dd_3:
ss.append(str(dd_3[sent][1]) + '_' + dd_3[sent][0])
pp_ss.append(dd_3[sent][0])
else:
if res[0][0] <= 0.5:
ss.append("0.0" + '_' + 'None')
pp_ss.append('None')
else:
ss.append(
str(res[0][0]) + '_' + id2label[res[1][0]])
pp_ss.append(id2label[res[1][0]])
if j == 0:
if res[0][0] <= 0.5:
ss.append("0.0" + '_' + 'None')
pp_ss.append('None')
else:
ss.append(str(res[0][0]) + '_' + id2label[res[1][0]])
pp_ss.append(id2label[res[1][0]])
pp_str = '_'.join(pp_ss).lower()
print(sent, '\t\t', pp_str)
if pp_str in stand_list:
fw.write("True" + '\t\t' + sent + '\t\t' + str(label) + '\n')
if int(label) == 3:
fw_false.write("True" + '\t\t' + sent + '\t\t' +
str(label) + '\n')
fw_false.write(" ".join(ss) + '\n')
fw_false.write('\n')
else:
if int(label) in [1, 2, 5]:
fw.write("False" + '\t\t' + sent + '\t\t' + str(label) +
'\n')
fw_false.write("False" + '\t\t' + sent + '\t\t' +
str(label) + '\n')
fw_false.write(" ".join(ss) + '\n')
fw_false.write('\n')
fw.write(" ".join(ss) + '\n')
fw.write('\n')
if int(label) == 3:
if pp_str in stand_list:
count_dict['3_correct'] = count_dict.get('3_correct',
0) + 1
else:
count_dict['3_error'] = count_dict.get('3_error', 0) + 1
if int(label) == 1:
if pp_str in stand_list:
count_dict['1_correct'] = count_dict.get('1_correct',
0) + 1
else:
count_dict['1_error'] = count_dict.get('1_error', 0) + 1
if int(label) == 2:
if pp_str in stand_list:
count_dict['2_correct'] = count_dict.get('2_correct',
0) + 1
else:
count_dict['2_error'] = count_dict.get('2_error', 0) + 1
if int(label) == 5:
if pp_str in stand_list:
count_dict['5_correct'] = count_dict.get('5_correct',
0) + 1
else:
count_dict['5_error'] = count_dict.get('5_error', 0) + 1
print(count_dict)