使用CNN进行文本分类代码实现

1、因为最近学习牛津大学的深度学习nlp课程，练习nlp文本分类，便进行了相应的调研，在课件中进行了全连接神经网络的文本分类，rnn文本分类。作者在之后也上网查看资料，了解到CNN在文本分类中发展很快，于是打算实现一个CNN文本分类的算法，查看资料，主要参考如下博文：http://www.wildml.com/2015/12/implementing-a-cnn-for-text-classification-in-tensorflow/

2、 作者讲的较为清晰且有相应github代码，但是完全tensorflow布置，感觉有很多东西模型使用起来不方便，于是学习keras或者更大受scikit-learn的影响，构造了一个相应的CNN文本分类器，相应的使用方法较为简单，使用时和scikit-learn一样仅仅需要三步（模型实体化，模型训练，模型预测）
相应代码如下（文件名为：TextCNNClassifier.py）：

# coding: utf-8
import tensorflow as tf
import numpy as np
import os

class NN_config(object):
    def __init__(self, vocab_size, num_filters,filter_steps,num_seqs=1000,num_classes=2, embedding_size=200):
        self.vocab_size     = vocab_size
        self.num_filters    = num_filters
        self.filter_steps   = filter_steps
        self.num_seqs       = num_seqs
        self.num_classes    = num_classes
        self.embedding_size = embedding_size        

class CALC_config(object):
    def __init__(self, learning_rate=0.0075, batch_size=64, num_epoches=20, l2_ratio=0.0):
        self.learning_rate = learning_rate
        self.batch_size    = batch_size
        self.num_epoches   = num_epoches
        self.l2_ratio      = l2_ratio

class TextCNNClassifier(object):
    '''
    A class used to define text classifier use convolution network 
    the form of class like keras or scikit-learn
    '''
    def __init__(self,config_nn, config_calc):

        self.num_seqs = config_nn.num_seqs
        self.num_classes = config_nn.num_classes
        self.embedding_size = config_nn.embedding_size
        self.vocab_size  = config_nn.vocab_size
        self.num_filters = config_nn.num_filters
        self.filter_steps = config_nn.filter_steps

        self.learning_rate = config_calc.learning_rate
        self.batch_size    = config_calc.batch_size
        self.num_epoches   = config_calc.num_epoches
        self.l2_ratio      = config_calc.l2_ratio

        #tf.reset_default_graph()
        self.build_placeholder()
        self.build_embedding_layer()
        self.build_nn()
        self.build_cost()
        self.build_optimizer()
        self.saver = tf.train.Saver()

    def build_placeholder(self):
        with tf.name_scope('inputs_to_data'):
            self.inputs       = tf.placeholder( tf.int32,shape=[None, self.num_seqs],name='inputs')
            self.targets      = tf.placeholder(tf.float32,shape=[None, self.num_classes], name='targets')
            self.keep_prob    = tf.placeholder(tf.float32, name='nn_keep_prob')
            print('self.inputs.shape:',self.inputs.shape)

    def build_embedding_layer(self):
        with tf.device('/cpu:0'),tf.name_scope('embeddings'):
            embeddings = tf.Variable(tf.truncated_normal(shape=[self.vocab_size,self.embedding_size],stddev=0.1),\
                        name = 'embeddings')
            x = tf.nn.embedding_lookup(embeddings, self.inputs)
            x = tf.expand_dims(x, axis=-1)
            self.x = tf.cast(x, tf.float32 )
            print('x shape is:',self.x.get_shape())

    def build_nn(self):
        conv_out = []
        for i , filter_step in enumerate(self.filter_steps):
            with tf.name_scope("conv-network-%s"%filter_step):
                filter_shape = [filter_step,self.embedding_size, 1,self.num_filters]
                filters = tf.Variable(tf.truncated_normal(s