TF2.3版本，如何tensor转numpy并可视化？

Question

#注意力
def attention_3d_block(inputs):
    #input_dim = int(inputs.shape[2])
    a = Permute((2, 1))(inputs)
    a = Dense(1, activation='softmax')(a)
    a_probs = Permute((2, 1), name='attention_vec')(a)
    #output_attention_mul = merge([inputs, a_probs], name='attention_mul', mode='mul')
    output_attention_mul = multiply([inputs, a_probs], name='attention_mul')
    #print('权重值000：', output_attention_mul)
    return output_attention_mul
def lstm(wavs,labels,testwavs,testlabels):
    feature = testwavs.shape[1]
    wavs=wavs.reshape(wavs.shape[0],1,4758)
    testwavs=testwavs.reshape(testwavs.shape[0],1,4758)
    cell=len(testlabels[1])
    cell=int(cell)
    inputs = Input(shape=(1, 4758))  # 输入特征接收维度
    drop = Dense(64, activation='relu')(inputs)  # 先进行全连接网络承接输入特征
    #lstm_out = Bidirectional(LSTM(64, return_sequences=True)(drop))  # 作为LSTM网络的输入
    lstm_out = LSTM(64, return_sequences=True)(drop) # 作为LSTM网络的输入
    attention_mul = attention_3d_block(lstm_out)  # 采用注意力机制

    print('权重值1：', attention_mul)
输出：权重值1： Tensor("attention_mul/Identity:0", shape=(None, 1, 64), dtype=float32)

    #attention_mul.numpy() 会报错：没有numpy module

    # attention_flatten = Flatten()(attention_mul)  # 展评特征向量维度
    attention_flatten = Flatten()(attention_mul)  # 展评特征向量维度
   
    # model.add(Dropout(0.2))
    drop2 = Dropout(0.8)(attention_flatten)  # 全连接
    drop3 = Dense(50, activation='sigmoid')(drop2)
    # model.add(Dense(64, activation='relu'))
    output = Dense(cell, activation='softmax')(drop3)  # 输出类别概率
    model = Model(inputs=inputs, outputs=output)  # 初始命名训练的模型为model
    # [编译模型] 配置模型，损失函数采用交叉熵，优化采用Adadelta，将识别准确率作为模型评估
    model.compile(loss='categorical_crossentropy', optimizer='adam',metrics=['accuracy'])#optimizer='adam', loss='binary_crossentropy', metrics=['accuracy']
    #  validation_data为验证集

Answer 1

不行。我都试了，听位朋友说代码段里input那里没有传入具体数据，所以没能计算出张量，我现在不知道怎么把数据传进去