现在很多卖货公司都使用聊天机器人充当客服人员，许多科技巨头也纷纷推出各自的聊天助手，如苹果Siri、Google Now、Amazon Alexa、微软小冰等等。前不久有一个视频比较了Google Now和Siri哪个更智能，貌似Google Now更智能。
本帖使用TensorFlow制作一个简单的聊天机器人。这个聊天机器人使用中文对话数据集进行训练（使用什么数据集训练决定了对话类型）。使用的模型为RNN(seq2seq)，和前文的《RNN生成古诗词》《RNN生成音乐》类似。
相关博文：
- 使用深度学习打造智能聊天机器人
- 脑洞大开：基于美剧字幕的聊天语料库建设方案
- 中文对白语料
- https://www.tensorflow.org/versions/r0.12/tutorials/seq2seq/index.html
- https://github.com/tflearn/tflearn/blob/master/examples/nlp/lstm_generator_shakespeare.py
数据集
我使用现成的影视对白数据集，跪谢作者分享数据。
下载数据集：
|
1
2
3
|
$ wget https://raw.githubusercontent.com/rustch3n/dgk_lost_conv/master/dgk_shooter_min.conv.zip
# 解压
$ unzip dgk_shooter_min.conv.zip
|
数据预处理：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
|
import os
import random
conv_path = 'dgk_shooter_min.conv'
if not os.path.exists(conv_path):
print('数据集不存在')
exit()
# 数据集格式
"""
E
M 畹/华/吾/侄/
M 你/接/到/这/封/信/的/时/候/
M 不/知/道/大/伯/还/在/不/在/人/世/了/
E
M 咱/们/梅/家/从/你/爷/爷/起/
M 就/一/直/小/心/翼/翼/地/唱/戏/
M 侍/奉/宫/廷/侍/奉/百/姓/
M 从/来/不/曾/遭/此/大/祸/
M 太/后/的/万/寿/节/谁/敢/不/穿/红/
M 就/你/胆/儿/大/
M 唉/这/我/舅/母/出/殡/
M 我/不/敢/穿/红/啊/
M 唉/呦/唉/呦/爷/
M 您/打/得/好/我/该/打/
M 就/因/为/没/穿/红/让/人/赏/咱/一/纸/枷/锁/
M 爷/您/别/给/我/戴/这/纸/枷/锁/呀/
E
M 您/多/打/我/几/下/不/就/得/了/吗/
M 走/
M 这/是/哪/一/出/啊/…/ / /这/是/
M 撕/破/一/点/就/弄/死/你/
M 唉/
M 记/着/唱/戏/的/再/红/
M 还/是/让/人/瞧/不/起/
M 大/伯/不/想/让/你/挨/了/打/
M 还/得/跟/人/家/说/打/得/好/
M 大/伯/不/想/让/你/再/戴/上/那/纸/枷/锁/
M 畹/华/开/开/门/哪/
E
...
"""
# 我首先使用文本编辑器sublime把dgk_shooter_min.conv文件编码转为UTF-8，一下子省了不少麻烦
convs = [] # 对话集合
with open(conv_path, encoding = "utf8") as f:
one_conv = [] # 一次完整对话
for line in f:
line = line.strip('\n').replace('/', '')
if line == '':
continue
if line[0] == 'E':
if one_conv:
convs.append(one_conv)
one_conv = []
elif line[0] == 'M':
one_conv.append(line.split(' ')[1])
"""
print(convs[:3]) # 个人感觉对白数据集有点不给力啊
[ ['畹华吾侄', '你接到这封信的时候', '不知道大伯还在不在人世了'],
['咱们梅家从你爷爷起', '就一直小心翼翼地唱戏', '侍奉宫廷侍奉百姓', '从来不曾遭此大祸', '太后的万寿节谁敢不穿红', '就你胆儿大', '唉这我舅母出殡', '我不敢穿红啊', '唉呦唉呦爷', '您打得好我该打', '就因为没穿红让人赏咱一纸枷锁', '爷您别给我戴这纸枷锁呀'],
['您多打我几下不就得了吗', '走', '这是哪一出啊 ', '撕破一点就弄死你', '唉', '记着唱戏的再红', '还是让人瞧不起', '大伯不想让你挨了打', '还得跟人家说打得好', '大伯不想让你再戴上那纸枷锁', '畹华开开门哪'], ....]
"""
# 把对话分成问与答
ask = [] # 问
response = [] # 答
for conv in convs:
if len(conv) == 1:
continue
if len(conv) % 2 != 0: # 奇数对话数, 转为偶数对话
conv = conv[:-1]
for i in range(len(conv)):
if i % 2 == 0:
ask.append(conv[i])
else:
response.append(conv[i])
"""
print(len(ask), len(response))
print(ask[:3])
print(response[:3])
['畹华吾侄', '咱们梅家从你爷爷起', '侍奉宫廷侍奉百姓']
['你接到这封信的时候', '就一直小心翼翼地唱戏', '从来不曾遭此大祸']
"""
def convert_seq2seq_files(questions, answers, TESTSET_SIZE = 8000):
# 创建文件
train_enc = open('train.enc','w') # 问
train_dec = open('train.dec','w') # 答
test_enc = open('test.enc', 'w') # 问
test_dec = open('test.dec', 'w') # 答
# 选择20000数据作为测试数据
test_index = random.sample([i for i in range(len(questions))],TESTSET_SIZE)
for i in range(len(questions)):
if i in test_index:
test_enc.write(questions[i]+'\n')
test_dec.write(answers[i]+ '\n' )
else:
train_enc.write(questions[i]+'\n')
train_dec.write(answers[i]+ '\n' )
if i % 1000 == 0:
print(len(range(len(questions))), '处理进度:', i)
train_enc.close()
train_dec.close()
test_enc.close()
test_dec.close()
convert_seq2seq_files(ask, response)
# 生成的*.enc文件保存了问题
# 生成的*.dec文件保存了回答
|
创建词汇表，然后把对话转为向量形式，参看练习1和7：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
|
# 前一步生成的问答文件路径
train_encode_file = 'train.enc'
train_decode_file = 'train.dec'
test_encode_file = 'test.enc'
test_decode_file = 'test.dec'
print('开始创建词汇表...')
# 特殊标记，用来填充标记对话
PAD = "__PAD__"
GO = "__GO__"
EOS = "__EOS__" # 对话结束
UNK = "__UNK__" # 标记未出现在词汇表中的字符
START_VOCABULART = [PAD, GO, EOS, UNK]
PAD_ID = 0
GO_ID = 1
EOS_ID = 2
UNK_ID = 3
# 参看tensorflow.models.rnn.translate.data_utils
vocabulary_size = 5000
# 生成词汇表文件
def gen_vocabulary_file(input_file, output_file):
vocabulary = {}
with open(input_file) as f:
counter = 0
for line in f:
counter += 1
tokens = [word for word in line.strip()]
for word in tokens:
if word in vocabulary:
vocabulary[word] += 1
else:
vocabulary[word] = 1
vocabulary_list = START_VOCABULART + sorted(vocabulary, key=vocabulary.get, reverse=True)
# 取前5000个常用汉字, 应该差不多够用了(额, 好多无用字符, 最好整理一下. 我就不整理了)
if len(vocabulary_list) > 5000:
vocabulary_list = vocabulary_list[:5000]
print(input_file + " 词汇表大小:", len(vocabulary_list))
with open(output_file, "w") as ff:
for word in vocabulary_list:
ff.write(word + "\n")
gen_vocabulary_file(train_encode_file, "train_encode_vocabulary")
gen_vocabulary_file(train_decode_file, "train_decode_vocabulary")
train_encode_vocabulary_file = 'train_encode_vocabulary'
train_decode_vocabulary_file = 'train_decode_vocabulary'
print("对话转向量...")
# 把对话字符串转为向量形式
def convert_to_vector(input_file, vocabulary_file, output_file):
tmp_vocab = []
with open(vocabulary_file, "r") as f:
tmp_vocab.extend(f.readlines())
tmp_vocab = [line.strip() for line in tmp_vocab]
vocab = dict([(x, y) for (y, x) in enumerate(tmp_vocab)])
#{'硕': 3142, 'v': 577, 'Ｉ': 4789, '\ue796': 4515, '拖': 1333, '疤': 2201 ...}
output_f = open(output_file, 'w')
with open(input_file, 'r') as f:
for line in f:
line_vec = []
for words in line.strip():
line_vec.append(vocab.get(words, UNK_ID))
output_f.write(" ".join([str(num) for num in line_vec]) + "\n")
output_f.close()
convert_to_vector(train_encode_file, train_encode_vocabulary_file, 'train_encode.vec')
convert_to_vector(train_decode_file, train_decode_vocabulary_file, 'train_decode.vec')
convert_to_vector(test_encode_file, train_encode_vocabulary_file, 'test_encode.vec')
convert_to_vector(test_decode_file, train_decode_vocabulary_file, 'test_decode.vec')
|
生成的train_encode.vec和train_decode.vec用于训练，对应的词汇表是train_encode_vocabulary和train_decode_vocabulary。
训练
需要很长时间训练，这还是小数据集，如果用百GB级的数据，没10天半个月也训练不完。
使用的模型：seq2seq_model.py。
代码：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
|
import tensorflow as tf # 0.12
from tensorflow.models.rnn.translate import seq2seq_model
import os
import numpy as np
import math
PAD_ID = 0
GO_ID = 1
EOS_ID = 2
UNK_ID = 3
train_encode_vec = 'train_encode.vec'
train_decode_vec = 'train_decode.vec'
test_encode_vec = 'test_encode.vec'
test_decode_vec = 'test_decode.vec'
# 词汇表大小5000
vocabulary_encode_size = 5000
vocabulary_decode_size = 5000
buckets = [(5, 10), (10, 15), (20, 25), (40, 50)]
layer_size = 256 # 每层大小
num_layers = 3 # 层数
batch_size = 64
# 读取*dencode.vec和*decode.vec数据（数据还不算太多, 一次读人到内存）
def read_data(source_path, target_path, max_size=None):
data_set = [[] for _ in buckets]
with tf.gfile.GFile(source_path, mode="r") as source_file:
with tf.gfile.GFile(target_path, mode="r") as target_file:
source, target = source_file.readline(), target_file.readline()
counter = 0
while source and target and (not max_size or counter < max_size):
counter += 1
source_ids = [int(x) for x in source.split()]
target_ids = [int(x) for x in target.split()]
target_ids.append(EOS_ID)
for bucket_id, (source_size, target_size) in enumerate(buckets):
if len(source_ids) < source_size and len(target_ids) < target_size:
data_set[bucket_id].append([source_ids, target_ids])
break
source, target = source_file.readline(), target_file.readline()
return data_set
model = seq2seq_model.Seq2SeqModel(source_vocab_size=vocabulary_encode_size, target_vocab_size=vocabulary_decode_size,
buckets=buckets, size=layer_size, num_layers=num_layers, max_gradient_norm= 5.0,
batch_size=batch_size, learning_rate=0.5, learning_rate_decay_factor=0.97, forward_only=False)
config = tf.ConfigProto()
config.gpu_options.allocator_type = 'BFC' # 防止 out of memory
with tf.Session(config=config) as sess:
# 恢复前一次训练
ckpt = tf.train.get_checkpoint_state('.')
if ckpt != None:
print(ckpt.model_checkpoint_path)
model.saver.restore(sess, ckpt.model_checkpoint_path)
else:
sess.run(tf.global_variables_initializer())
train_set = read_data(train_encode_vec, train_decode_vec)
test_set = read_data(test_encode_vec, test_decode_vec)
train_bucket_sizes = [len(train_set[b]) for b in range(len(buckets))]
train_total_size = float(sum(train_bucket_sizes))
train_buckets_scale = [sum(train_bucket_sizes[:i + 1]) / train_total_size for i in range(len(train_bucket_sizes))]
loss = 0.0
total_step = 0
previous_losses = []
# 一直训练，每过一段时间保存一次模型
while True:
random_number_01 = np.random.random_sample()
bucket_id = min([i for i in range(len(train_buckets_scale)) if train_buckets_scale[i] > random_number_01])
encoder_inputs, decoder_inputs, target_weights = model.get_batch(train_set, bucket_id)
_, step_loss, _ = model.step(sess, encoder_inputs, decoder_inputs, target_weights, bucket_id, False)
loss += step_loss / 500
total_step += 1
print(total_step)
if total_step % 500 == 0:
print(model.global_step.eval(), model.learning_rate.eval(), loss)
# 如果模型没有得到提升，减小learning rate
if len(previous_losses) > 2 and loss > max(previous_losses[-3:]):
sess.run(model.learning_rate_decay_op)
previous_losses.append(loss)
# 保存模型
checkpoint_path = "chatbot_seq2seq.ckpt"
model.saver.save(sess, checkpoint_path, global_step=model.global_step)
loss = 0.0
# 使用测试数据评估模型
for bucket_id in range(len(buckets)):
if len(test_set[bucket_id]) == 0:
continue
encoder_inputs, decoder_inputs, target_weights = model.get_batch(test_set, bucket_id)
_, eval_loss, _ = model.step(sess, encoder_inputs, decoder_inputs, target_weights, bucket_id, True)
eval_ppx = math.exp(eval_loss) if eval_loss < 300 else float('inf')
print(bucket_id, eval_ppx)
|
聊天机器人
使用训练好的模型：
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
|
import tensorflow as tf # 0.12
from tensorflow.models.rnn.translate import seq2seq_model
import os
import numpy as np
PAD_ID = 0
GO_ID = 1
EOS_ID = 2
UNK_ID = 3
train_encode_vocabulary = 'train_encode_vocabulary'
train_decode_vocabulary = 'train_decode_vocabulary'
def read_vocabulary(input_file):
tmp_vocab = []
with open(input_file, "r") as f:
tmp_vocab.extend(f.readlines())
tmp_vocab = [line.strip() for line in tmp_vocab]
vocab = dict([(x, y) for (y, x) in enumerate(tmp_vocab)])
return vocab, tmp_vocab
vocab_en, _, = read_vocabulary(train_encode_vocabulary)
_, vocab_de, = read_vocabulary(train_decode_vocabulary)
# 词汇表大小5000
vocabulary_encode_size = 5000
vocabulary_decode_size = 5000
buckets = [(5, 10), (10, 15), (20, 25), (40, 50)]
layer_size = 256 # 每层大小
num_layers = 3 # 层数
batch_size = 1
model = seq2seq_model.Seq2SeqModel(source_vocab_size=vocabulary_encode_size, target_vocab_size=vocabulary_decode_size,
buckets=buckets, size=layer_size, num_layers=num_layers, max_gradient_norm= 5.0,
batch_size=batch_size, learning_rate=0.5, learning_rate_decay_factor=0.99, forward_only=True)
model.batch_size = 1
with tf.Session() as sess:
# 恢复前一次训练
ckpt = tf.train.get_checkpoint_state('.')
if ckpt != None:
print(ckpt.model_checkpoint_path)
model.saver.restore(sess, ckpt.model_checkpoint_path)
else:
print("没找到模型")
while True:
input_string = input('me > ')
# 退出
if input_string == 'quit':
exit()
input_string_vec = []
for words in input_string.strip():
input_string_vec.append(vocab_en.get(words, UNK_ID))
bucket_id = min([b for b in range(len(buckets)) if buckets[b][0] > len(input_string_vec)])
encoder_inputs, decoder_inputs, target_weights = model.get_batch({bucket_id: [(input_string_vec, [])]}, bucket_id)
_, _, output_logits = model.step(sess, encoder_inputs, decoder_inputs, target_weights, bucket_id, True)
outputs = [int(np.argmax(logit, axis=1)) for logit in output_logits]
if EOS_ID in outputs:
outputs = outputs[:outputs.index(EOS_ID)]
response = "".join([tf.compat.as_str(vocab_de[output]) for output in outputs])
print('AI > ' + response)
|
测试
额，好差劲。
上面的实现并没有用到任何自然语言的特性（分词、语法等等），只是单纯的使用数据强行提高它的“智商”。
后续练习：中文语音识别、文本转语音
http://blog.topspeedsnail.com/archives/10735
使用谷歌开源的TensorFlow进行一系列的训练实践 https://www.urlteam.org
tensorflow
使用谷歌开源的TensorFlow进行一系列的训练实践
2017.9.10 重启项目
项目列表
前三篇主要学习自熊猫的博客：http://blog.topspeedsnail.com/
1:Cnn_Captcha
用卷积神经网络识别复杂字符验证码
用4层Cnn网络，识别破解python自生成的复杂扭曲验证码，
项目实践说明：https://zhuanlan.zhihu.com/p/25779608
2:Forecast
用CNN根据名字判断性别
3.Rnn_Create_Poetry
用RNN生成古诗词
4.Object_Detection,目标检测
本项目是使用tensorflow的Object Detection API，进行目标检测。
完成的是数据预处理脚本，如何用自己的数据生成可以训练的record文件。
https://github.com/luyishisi/tensorflow