练习：NLP之文本分类---(1)Word2Vec+Conv2d+AG_news

目录

一、任务

二、思路

三、数据预处理

四、建立模型

五、模型结果

六、报错及处理

1.Error： 'Tensor' object has no attribute 'lower'

2.Error: ValueError: The last dimension of the inputs to `Dense` should be defined. Found `None`

---

一、任务

1. 数据为AG_news 新闻数据集标题文本
2. word2vec进行词向量训练 并输入卷积模型
3. 利用多种卷积核3*64、5*64、7*64来训练
4. 最大池化转二维结构
5. 将池化结果连接
6. 全连接层分类---5类

二、思路

随手画的---后面模型按这个思路 单数换了参数 懒得画了 将就一下

三、数据预处理

1. 标题文本清洗
2. 标签one-hot
3. 标题文本词向量word2vec 向量大小为12 不够补齐 多余截断

"""一、单词文本处理函数"""
#1.文本清理函数（大小写、去除标点、删除多余空格、首尾空格删除、以空格分词

def text_clearTitle_word2vec(text):
    text = text.lower()  					#将文本转化成小写
    text = re.sub(r"[^a-z]"," ",text)   	#替换非标准字符，^是求反操作。
    text = re.sub(r" +", " ", text) 		#替换多重空格
    text = text.strip()  					#取出首尾空格
    text = text + " eos"                 	#添加结束符
    text = text.split(" ")
    return text

#2.将标签转为one-hot格式函数
def get_label_one_hot(list):
    values = np.array(list)
    # print("values:",values)
    n_values = np.max(values) + 1
    # print(np.max(values))
    return np.eye(n_values)[values]

#3.获取训练集和标签函数
def get_word2vec_dataset(n = 12):
    #读取数据并完成预处理
    agnews_label = []
    agnews_title = []
    agnews_train = csv.reader(open("./dataset/train.csv", "r"))
    for line in agnews_train:
        agnews_label.append(np.int(line[0]))
        agnews_title.append(text_clearTitle_word2vec(line[1]))

    #建立word2vec模型
    model = word2vec.Word2Vec(agnews_title,vector_size=64,min_count=0,window=5)

    train_dataset = []
    for line in agnews_title:
        length = len(line)
        if length > n:
            line = line[:n]
            word2vec_matix = (model.wv[line])
            train_dataset.append(word2vec_matix)
        else:
            word2vec_matix = (model.wv[line])
            pad_length = n - length
            pad_matrix = np.zeros([pad_length,64])+1e-10
            word2vec_matix = np.concatenate([word2vec_matix,pad_matrix],axis=0)
            train_dataset.append(word2vec_matix)
    train_dataset = np.expand_dims(train_dataset,3)
    print(train_dataset.shape)
    print(train_dataset.ndim)

    label_dataset = get_label_one_hot(agnews_label)
    return train_dataset,label_dataset

四、建立模型

"""二、创建word2vec模型"""
#word2vec_CNN的模型
def word2vec_CNN():
    xs = tf.keras.Input(shape=(12,64,1))
    #设置卷积核大小为[3,64]通道为12的卷积计算
    conv_3 = tf.keras.layers.Conv2D(12,[3,64],activation=tf.nn.relu)(xs)
    #设置卷积核大小为[5,64]通道为12的卷积计算
    conv_5 = tf.keras.layers.Conv2D(12,[5,64],activation=tf.nn.relu)(xs)
    #设置卷积核大小为[7,64]通道为12的卷积计算
    conv_7 = tf.keras.layers.Conv2D(12,[7,64],activation=tf.nn.relu)(xs)
    #下面是分别对卷积计算的结果进行池化处理，将池化处理的结果转成二维结构
    conv_3_mean = tf.keras.layers.Flatten()(tf.reduce_max(conv_3,axis=1,keepdims=True))
    conv_5_mean = tf.keras.layers.Flatten()(tf.reduce_max(conv_5,axis=1,keepdims=True))
    conv_7_mean = tf.keras.layers.Flatten()(tf.reduce_max(conv_7,axis=1,keepdims=True))

    flatten = tf.concat([conv_3_mean,conv_5_mean,conv_7_mean],axis=1)
    fc_1 = tf.keras.layers.Dense(128,activation=tf.nn.relu)(flatten)
    logits = tf.keras.layers.Dense(5,activation=tf.nn.softmax)(fc_1)
    model = tf.keras.Model(inputs = xs, outputs = logits)
    return model

train_dataset,label_dataset = get_word2vec_dataset()
X_train,X_test,y_train,y_test = train_test_split(train_dataset,label_dataset,test_size=0.1,random_state=217)

batch_size = 12
train_data = tf.data.Dataset.from_tensor_slices((X_train,y_train)).batch(batch_size)
model = word2vec_CNN()
model.compile(optimizer = tf.optimizers.Adam(1e-3),loss = tf.losses.categorical_crossentropy,metrics=['accuracy'])
model.fit(train_data,epochs=10)

五、模型结果

1.运行结果

score = model.evaluate(X_test,y_test)
print("score:",score)

2.可视化

history = model.fit(train_data,epochs=10)
history_dict = history.history
train_loss = history_dict["loss"]
train_accuracy = history_dict["accuracy"]


plt.figure()
plt.plot(range(10), train_loss, label='train_loss')
plt.legend()
plt.plot(range(10), train_accuracy, label='train_accuracy')
plt.legend()
plt.xlabel('epochs')
plt.ylabel('loss & accuracy')
plt.show()

六、报错及处理

1.Error： 'Tensor' object has no attribute 'lower'

1）前后代码：

    xs = tf.keras.Input(shape=(12,64,1))
    #设置卷积核大小为[3,64]通道为12的卷积计算
    conv_3 = tf.keras.layers.Conv2D(12,[3,64],activation=tf.nn.relu)(xs)
    #设置卷积核大小为[5,64]通道为12的卷积计算
    conv_5 = tf.keras.layers.Conv2D(12,[5,64],activation=tf.nn.relu)(xs)
    #设置卷积核大小为[7,64]通道为12的卷积计算
    conv_7 = tf.keras.layers.Conv2D(12,[7,64],activation=tf.nn.relu)(xs)
    #下面是分别对卷积计算的结果进行池化处理，将池化处理的结果转成二维结构
    conv_3_mean = tf.keras.layers.Flatten()(tf.reduce_max(conv_3,axis=1,keepdims=True))
    conv_5_mean = tf.keras.layers.Flatten()(tf.reduce_max(conv_5,axis=1,keepdims=True))
    conv_7_mean = tf.keras.layers.Flatten()(tf.reduce_max(conv_7,axis=1,keepdims=True))

2）报错行：

conv_3_mean = tf.keras.layers.Flatten(tf.reduce_max(conv_3,axis=1,keepdims=True))

3）修改：

 参考：python - 'Tensor' object has no attribute 'lower' - Stack Overflow

conv_3_mean = tf.keras.layers.Flatten()(tf.reduce_max(conv_3,axis=1,keepdims=True))

2.Error: ValueError: The last dimension of the inputs to `Dense` should be defined. Found `None`

1）问题：Dense层在输入的时候，需要定义shape/size

2）解决：在tf.keras.Input层定义shape---不要用None

# xs = tf.keras.Input([None, None])
xs = tf.keras.Input(shape=(12,64,1))