import tensorflow as tf
tf.__version__
'2.2.2'
import keras
print(keras.__version__)
2.3.1
Using TensorFlow backend.
from tensorflow import keras
print(keras.__version__)
2.3.0-tf
from tensorflow.keras.utils import to_categorical
from sklearn.model_selection import train_test_split
MAX_LENGTH = 1024
#载入训练集和测试集数据
# train = pd.read_csv(r'D:\code\python\csdn_nlp\data\文本分类数据\Texts_Classification\train2.csv')
# test = pd.read_csv(r'D:\code\python\csdn_nlp\data\文本分类数据\Texts_Classification\test2.csv')
train = pd.read_csv(r'/mnt/d/code/python/csdn_nlp/data/文本分类数据/Texts_Classification/train2.csv')
test = pd.read_csv(r'/mnt/d/code/python/csdn_nlp/data/文本分类数据/Texts_Classification/test2.csv')
train.head(10)
| Unnamed: 0 | 分类 | 文章 | 字数 | 分词 | label |
|---|
| 0 | 10000 | 体育 | 欧洲杯观察：足坛格局地震 东南欧已能占半边天 搜狐体育讯 土耳其与克罗地亚在本届欧洲杯１／４... | 1818 | ['欧洲杯', '观察', '足坛', '格局', '地震', '东南欧', '已能', '... | 8 |
|---|
| 1 | 10001 | 体育 | 来源：搜狐体育 作者：雷欧 主队 比分 客队 进球视频青岛 １－１ 武汉浙江 １－１ 北京搜... | 2190 | ['来源', '搜狐', '体育', '作者', '雷欧', '主队', '比分', '客队... | 8 |
|---|
| 2 | 10002 | 体育 | 欧洲杯－帕夫柳琴科范尼破门 荷兰１－１俄罗斯加时 搜狐体育讯 北京时间６月２２日，２００８欧... | 2078 | ['欧洲杯', '帕夫', '柳琴', '科范尼', '破门', '荷兰', '俄罗斯', ... | 8 |
|---|
| 3 | 10003 | 体育 | 郅联璧合配合日臻完善 斯杯之前尤纳斯再谈裁员 ７年后重登上海滩，中国男篮来去匆匆。队长刘炜回... | 1810 | ['郅', '联璧', '合', '配合', '日臻完善', '斯杯', '之前', '尤纳... | 8 |
|---|
| 4 | 10004 | 体育 | 盘踞世界第一逼近五百周 伍兹创造高尔夫里程碑 搜狐体育讯 下周一，老虎－伍兹将第４９９周稳坐... | 1893 | ['盘踞', '世界', '第一', '逼近', '五百', '周伍', '兹', '创造'... | 8 |
|---|
| 5 | 10005 | 体育 | 第五日看点：郑洁挑战ＮＯ．１ 费天王冲击男单１６强 搜狐体育讯 北京时间６月２７日晚，２００... | 2082 | ['第五日', '看点', '郑洁', '挑战', '费', '天王', '冲击', '男单... | 8 |
|---|
| 6 | 10007 | 体育 | 王德显改变双姝命运 邢慧娜孙英杰成败均因一人 ２００７年年底，邢慧娜来到了北京。田管中心已经... | 1855 | ['王德显', '改变', '双姝', '命运', '邢慧娜', '孙', '英杰', '成... | 8 |
|---|
| 7 | 10008 | 体育 | 半场实录：皮尔斯异军突起 湖人队领先凯尔特人 搜狐直播员：搜狐网友：赛式是２３２制的，还是主... | 8550 | ['半场', '实录', '皮尔斯', '异军突起', '湖人队', '领先', '凯尔特人... | 8 |
|---|
| 8 | 10009 | 体育 | 温格：德国可成英格兰榜样 三狮军团差在意志力 晨报特派记者 甘慧（奥地利维也纳６月２９日电）... | 2162 | ['温格', '德国', '可成', '英格兰', '榜样', '三狮', '军团', '差... | 8 |
|---|
| 9 | 10010 | 体育 | 欧洲杯－范尼斯内德建功 荷兰３－０狂屠意大利 搜狐体育讯 北京时间６月１０日凌晨２：４５，２... | 2643 | ['欧洲杯', '范', '尼斯', '德', '建功', '荷兰', '狂屠', '意大利... | 8 |
|---|
temp_train_data_list = train[["文章","label"]].values.tolist()
train_data_list = [tuple(x) for x in temp_train_data_list]
temp_test_data_list = test[["文章","label"]].values.tolist()
test_data_list = [tuple(x) for x in temp_test_data_list]
train_data_list[0]
print(type(train_data_list))
<class 'list'>
# from sklearn.model_selection import train_test_split
# '''不采取分层抽样时的数据集分割'''
# valid_data_list = train_test_split(train_data_list,test_size=0.3)
# len(valid_data_list)
# valid_data_list[0]
import random
baifenbi = 80
length = len(train_data_list)
print(length)
len(train_data_list[int((length/100)*baifenbi):])
23549
4710
# 加载数据集 这么取 值 标签训练会有问题,所以要随机打乱,然后选取
import random
random.shuffle(train_data_list)
train_data = train_data_list[0:int((length/100)*baifenbi)]
valid_data = train_data_list[int((length/100)*baifenbi):]
# random.seed(10)
# valid_data = random.sample(train_data_list, 5)
test_data = test_data_list
import numpy as np
from sklearn import metrics
from bert4keras.tokenizers import Tokenizer
from bert4keras.backend import keras, set_gelu
from bert4keras.models import build_transformer_model
from bert4keras.snippets import DataGenerator, sequence_padding
from bert4keras.optimizers import Adam, extend_with_piecewise_linear_lr
from keras.layers import Lambda, Dense
from contextlib import redirect_stdout
set_gelu('tanh') # 切换gelu版本
# 文本共计 12 个类别
num_classes = 12
# 长文本,长一些,好像bert 类,最长 512? 所以编码的文字 应该去掉停用词
maxlen = 64
batch_size = 8
# config_path = r"D:/code/python/csdn_nlp/model/albert_xlarge_zh/albert_xlarge/albert_config.json"
# checkpoint_path =tf.train.latest_checkpoint(r"D:/code/python/csdn_nlp/model/albert_xlarge_zh/albert_xlarge/")
# #checkpoint_path = r"D:/code/python/csdn_nlp/model/albert_xlarge_zh/albert_xlarge/model"
# dict_path = r"D:/code/python/csdn_nlp/model/albert_xlarge_zh/albert_xlarge/vocab_chinese.txt"
# linux 路径
config_path = r"/mnt/d/code/python/csdn_nlp/model/albert_base/albert_config.json"
# albert-base-v2 的压缩包里面没有ckpt，windows 下面加载报错，所以应该在linux 下面使用
checkpoint_path =tf.train.latest_checkpoint(r"/mnt/d/code/python/csdn_nlp/model/albert_base/model.ckpt-best.data-00000-of-00001")
#checkpoint_path = r"/mnt/d/code/python/csdn_nlp/model/albert_base/"
dict_path = r"/mnt/d/code/python/csdn_nlp/model/albert_base/vocab_chinese.txt"
# 建立分词器
tokenizer = Tokenizer(dict_path, do_lower_case=True)
class data_generator(DataGenerator):
"""数据生成器
"""
def __iter__(self, random=False):
batch_token_ids, batch_segment_ids, batch_labels = [], [], []
for is_end, (text, label) in self.sample(random):
token_ids, segment_ids = tokenizer.encode(text, maxlen=maxlen)
batch_token_ids.append(token_ids)
batch_segment_ids.append(segment_ids)
batch_labels.append([label])
if len(batch_token_ids) == self.batch_size or is_end:
batch_token_ids = sequence_padding(batch_token_ids)
batch_segment_ids = sequence_padding(batch_segment_ids)
batch_labels = sequence_padding(batch_labels)
yield [batch_token_ids, batch_segment_ids], batch_labels
batch_token_ids, batch_segment_ids, batch_labels = [], [], []
# 加载预训练模型
bert = build_transformer_model(
config_path=config_path,
checkpoint_path=checkpoint_path,
model='albert',
return_keras_model=False,
)
output = Lambda(lambda x: x[:, 0], name='CLS-token')(bert.model.output)
output = Dense(
units=num_classes,
activation='softmax',
kernel_initializer=bert.initializer
)(output)
model = keras.models.Model(bert.model.input, output)
model.summary()
# 派生为带分段线性学习率的优化器。
# 其中name参数可选，但最好填入，以区分不同的派生优化器。
AdamLR = extend_with_piecewise_linear_lr(Adam, name='AdamLR')
model.compile(
loss='sparse_categorical_crossentropy',
# optimizer=Adam(1e-5), # 用足够小的学习率
optimizer=AdamLR(learning_rate=1e-4, lr_schedule={
1000: 1,
2000: 0.1
}),
metrics=['accuracy'],
)
# 转换数据集
train_generator = data_generator(train_data, batch_size)
valid_generator = data_generator(valid_data, batch_size)
test_generator = data_generator(test_data, batch_size)
def evaluate(data):
total, right = 0., 0.
for x_true, y_true in data:
y_pred = model.predict(x_true).argmax(axis=1)
y_true = y_true[:, 0]
total += len(y_true)
right += (y_true == y_pred).sum()
return right / total
class Evaluator(keras.callbacks.Callback):
"""评估与保存
"""
def __init__(self):
self.best_val_acc = 0.
def on_epoch_end(self, epoch, logs=None):
val_acc = evaluate(valid_generator)
if val_acc > self.best_val_acc:
self.best_val_acc = val_acc
model.save_weights('best_model.weights')
test_acc = evaluate(test_generator)
print(
u'val_acc: %.5f, best_val_acc: %.5f, test_acc: %.5f\n' %
(val_acc, self.best_val_acc, test_acc)
)
Model: "model_6"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
Input-Token (InputLayer) (None, None) 0
__________________________________________________________________________________________________
Input-Segment (InputLayer) (None, None) 0
__________________________________________________________________________________________________
Embedding-Token (Embedding) (None, None, 128) 2704384 Input-Token[0][0]
__________________________________________________________________________________________________
Embedding-Segment (Embedding) (None, None, 128) 256 Input-Segment[0][0]
__________________________________________________________________________________________________
Embedding-Token-Segment (Add) (None, None, 128) 0 Embedding-Token[0][0]
Embedding-Segment[0][0]
__________________________________________________________________________________________________
Embedding-Position (PositionEmb (None, None, 128) 65536 Embedding-Token-Segment[0][0]
__________________________________________________________________________________________________
Embedding-Norm (LayerNormalizat (None, None, 128) 256 Embedding-Position[0][0]
__________________________________________________________________________________________________
Embedding-Mapping (Dense) (None, None, 768) 99072 Embedding-Norm[0][0]
__________________________________________________________________________________________________
Transformer-MultiHeadSelfAttent (None, None, 768) 2362368 Embedding-Mapping[0][0]
Embedding-Mapping[0][0]
Embedding-Mapping[0][0]
Transformer-FeedForward-Norm[0][0
Transformer-FeedForward-Norm[0][0
Transformer-FeedForward-Norm[0][0
Transformer-FeedForward-Norm[1][0
Transformer-FeedForward-Norm[1][0
Transformer-FeedForward-Norm[1][0
Transformer-FeedForward-Norm[2][0
Transformer-FeedForward-Norm[2][0
Transformer-FeedForward-Norm[2][0
Transformer-FeedForward-Norm[3][0
Transformer-FeedForward-Norm[3][0
Transformer-FeedForward-Norm[3][0
Transformer-FeedForward-Norm[4][0
Transformer-FeedForward-Norm[4][0
Transformer-FeedForward-Norm[4][0
Transformer-FeedForward-Norm[5][0
Transformer-FeedForward-Norm[5][0
Transformer-FeedForward-Norm[5][0
Transformer-FeedForward-Norm[6][0
Transformer-FeedForward-Norm[6][0
Transformer-FeedForward-Norm[6][0
Transformer-FeedForward-Norm[7][0
Transformer-FeedForward-Norm[7][0
Transformer-FeedForward-Norm[7][0
Transformer-FeedForward-Norm[8][0
Transformer-FeedForward-Norm[8][0
Transformer-FeedForward-Norm[8][0
Transformer-FeedForward-Norm[9][0
Transformer-FeedForward-Norm[9][0
Transformer-FeedForward-Norm[9][0
Transformer-FeedForward-Norm[10][
Transformer-FeedForward-Norm[10][
Transformer-FeedForward-Norm[10][
__________________________________________________________________________________________________
Transformer-MultiHeadSelfAttent (None, None, 768) 0 Embedding-Mapping[0][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[0][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[1][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[2][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[3][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[4][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[5][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[6][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[7][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[8][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[9][0
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward-Norm[10][
Transformer-MultiHeadSelfAttentio
__________________________________________________________________________________________________
Transformer-MultiHeadSelfAttent (None, None, 768) 1536 Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
__________________________________________________________________________________________________
Transformer-FeedForward (FeedFo (None, None, 768) 4722432 Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
Transformer-MultiHeadSelfAttentio
__________________________________________________________________________________________________
Transformer-FeedForward-Add (Ad (None, None, 768) 0 Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[0][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[1][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[2][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[3][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[4][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[5][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[6][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[7][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[8][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[9][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[10][0]
Transformer-MultiHeadSelfAttentio
Transformer-FeedForward[11][0]
__________________________________________________________________________________________________
Transformer-FeedForward-Norm (L (None, None, 768) 1536 Transformer-FeedForward-Add[0][0]
Transformer-FeedForward-Add[1][0]
Transformer-FeedForward-Add[2][0]
Transformer-FeedForward-Add[3][0]
Transformer-FeedForward-Add[4][0]
Transformer-FeedForward-Add[5][0]
Transformer-FeedForward-Add[6][0]
Transformer-FeedForward-Add[7][0]
Transformer-FeedForward-Add[8][0]
Transformer-FeedForward-Add[9][0]
Transformer-FeedForward-Add[10][0
Transformer-FeedForward-Add[11][0
__________________________________________________________________________________________________
CLS-token (Lambda) (None, 768) 0 Transformer-FeedForward-Norm[11][
__________________________________________________________________________________________________
dense_21 (Dense) (None, 12) 9228 CLS-token[0][0]
==================================================================================================
Total params: 9,966,604
Trainable params: 9,966,604
Non-trainable params: 0
__________________________________________________________________________________________________
evaluator = Evaluator()
model.fit(
train_generator.forfit(),
steps_per_epoch=len(train_generator),
epochs=1,
callbacks=[evaluator]
)
model.load_weights('best_model.weights')
print(u'final test acc: %05f\n' % (evaluate(test_generator)))
Epoch 1/1
2355/2355 [==============================] - 1663s 706ms/step - loss: 2.5211 - accuracy: 0.0876
val_acc: 0.08493, best_val_acc: 0.08493, test_acc: 0.08333
final test acc: 0.083333
