tf.expand_dims(
    input,
    axis,
    name=None
)

class Decoder(keras.Model):#使用子类API实现
    def __init__(self, vocab_size, embedding_units, decoding_units, batch_size):
        super(Decoder, self).__init__()
        self.batch_size = batch_size
        self.decoding_units = decoding_units
        self.embedding = keras.layers.Embedding(vocab_size, embedding_units) # 每一个单词转换成固定的向量，输出是三维的。
        self.gru = keras.layers.GRU(self.decoding_units, return_sequences=True, return_state=True, recurrent_initializer='glorot_uniform')
        self.fc = keras.layers.Dense(vocab_size)
        self.attention = BahdanauAttention(self.decoding_units)

        
        
    def call(self, x, hidden, encoding_outputs): #x是decoder中当前步的输入，hidden decoder中当前步的上一步的输出，encoding_outputs encoder中当前步的输出
	    # call 函数可以理解为对构造函数进行初始化。
		# 先计算hidden encoding_outputs 的attention
        
        # contex_vector.shape=(batch_size, units)
        
        context_vector, attention_weights = self.attention(hidden, encoding_outputs)
        print("正在计算Decoder-------------------------------------------------")
        print(tf.shape(x))
        x = self.embedding(x) # x 需要先进入embedding层， 目的是为了编码。会增加维度。在电信号上是不需要的。我使用的是expand_dim()进行维度的扩展。
        
        # 把context_vector和x 相加，这样就会携带每一步的输入，
        combined_x = tf.concat([tf.expand_dims(context_vector,1),x], axis=-1)
        
        # 这一步就是decoder的返回值，和encode类似，就是outout和hidden
#         output.shape = [batcha_size, 1 , decode_units]
#         state.shape = [batch_size, decode_units ]
        output, state = self.gru(combined_x)
        
        # 这里对output进行reshape       
        # reshape后 output.shape=[batch_size, decoder_units]
        output = tf.reshape(output, (-1, output.shape[2]))
        
        # 输入到全链接层，输出为[batch_size,vocab_sie]
        output = self.fc(output)
        
        return output, state, attention_weights
    
    
decoder = Decoder(output_vocab_size, embedding_units, units, batch_size)

outputs = decoder(tf.random.uniform((batch_size,1)), sample_hidden, sample_output)

decoder_output, decoder_hidden, decoder_aw = outputs

print(decoder_output.shape, decoder_hidden.shape, decoder_aw.shape)

indices[0,0] = -13 is not in [0, 600) [Op:ResourceGather] name: encoder_3/embedding_2/embedding_lookup

unsigned char xdata hc[3]={0,0,0};


//-----------------存储标定的k、b值--------------
void Save_flash(void)
{
    uint8 i,j;
    EA=0;              //禁止中断
    FLSCL=0x89;        //FLASH 写/擦除定时(Fosc=18.432MHz) 允许FLASH写
    PSCTL=0x07;        //临时FLASH 允许FLASH 写/擦除 
    FLASH_ADDR[0]=0;   //向待擦除扇区内的任何一个地址写入一个数据字节？擦除 临时FLASH
    for(i=0;i<2;i++)  //写入数据FLASH
    {
        PSCTL=0x00;    //程序数据FLASH 禁止FLASH 写/擦除 
	    j=hc[i];     //读FLASH存储器 与PSWE无关
        PSCTL=0x05;    //临时FLASH 禁止擦除 允许写 
	    FLASH_ADDR[i]=j;
    }                 
    FLSCL=0x4f;        //禁止FLASH 写/擦除
    PSCTL=0x00;        //程序数据FLASH 禁止FLASH 写/擦除 
    EA=1;
}
//-------------从MCU中读取出标定的k、b值------------------------
void Read_flash(void)        
{
    uint8 i,j;
	uint8 code *FLASH_READ;   //code定义的数据在flash中
	FLASH_READ=0x0000;
	for(i=0;i<2;i++)
	{
	    PSCTL=0x04;           //选择数据FLASH扇区 禁止FLASH 写/擦除
		j=FLASH_READ[i];      //读取数据flash 0x0000区
	    PSCTL=0x00;           //选择程序Flash 禁止写/擦除
		hc[i]=j;            //读程序flash
	}
}


void main()
{
    
	WDTCN = 0xde;                       // Disable watchdog timer  关闭看门狗
   	WDTCN = 0xad; 
 	SYSCLK_Init();
 	PORT_Init();
	ADC0_Init();	
	UART0_Init();
	
	Read_All();
	Para[0]++;
	Save_Para();
	
	
	P2 = 0x00;
	EA = 1;                             // Enable global interrupts
	
	Send_Buf0[0] = Para[0];    //发送帧头
	Send_Buf0[1] = 0x00; 
	Send_Max = 2;
	Send_p = 0;
	SCON0 &= 0xef; //REN0 = 0 禁止开始接收
	SBUF0  = 0x41; //启动回答数据事件

	while(1)											 
	{		
 	WDTCN = 0xa5;	  //使能并复位看门狗定时器
	}
}