我有一个用于情感识别的多模态模型，并对6类情感进行分类模型的结构是每个模态（音频，文本）的两个lstm，我想获得这种模型的混淆矩阵，classification_report和precision_recall_fscore_suppor，这是代码

if self.modality == "bimodal_":
                self.train_x_text  = self.train_x[:,:,:50].astype('int64')
                self.train_x_audio = self.train_x[:,:,50:]
                self.val_x_text  = self.val_x[:,:,:50].astype('int64')
                self.val_x_audio = self.val_x[:,:,50:]
                self.test_x_text  = self.test_x[:,:,:50].astype('int64')
                self.test_x_audio = self.test_x[:,:,50:]
def get_bimodal_model_(self):

            # Modality specific hyperparameters
            self.epochs = 90
            self.batch_size = 80

            # text modality
            
            scaler = MinMaxScaler()
            min_vals = np.min(self.train_x_text)
            max_vals = np.max(self.train_x_text)
            self.train_x_text = (self.train_x_text - min_vals) / (max_vals - min_vals)
            self.sentence_length = self.train_x_text.shape[2] # 50

            text_input  = Input(shape=(self.sequence_length, self.sentence_length), dtype='float32')
            cnn_outputs = self.cnn_text_feature_extractor(text_input) # return (33, 100)
            lstm_text = Bidirectional(LSTM(500, return_sequences=True,dropout=0.7, recurrent_dropout=0.8))(cnn_outputs)
            lstm_text= Dense(128, activation='relu')(lstm_text )
            
            lstm_text = BatchNormalization()(lstm_text)
            
                    
            # audio modality 
            min_vals_a = np.min(self.train_x_audio)
            max_vals_a = np.max(self.train_x_audio)
            self.train_x_audio = (self.train_x_audio - min_vals_a) / (max_vals_a - min_vals_a)
            self.features_audio_dim = self.train_x_audio.shape[2] #1611 
            audio_input  = Input(shape=(self.sequence_length, self.features_audio_dim), dtype='float32')
            lstm_audio = Bidirectional(LSTM(500, return_sequences=True,dropout=0.3, recurrent_dropout=0.8))(audio_input)
            lstm_audio= Dense(128, activation='relu')(lstm_audio )
            
            lstm_audio = BatchNormalization()(lstm_audio)
            concatenated_output = Concatenate()([lstm_text, lstm_audio])
            concatenated_output =Bidirectional(LSTM(500, return_sequences=True,dropout=0.3, recurrent_dropout=0.8))(concatenated_output)
            
            attention = Attention()([concatenated_output, concatenated_output])
            dense_1 = Dense(128, activation='sigmoid')(attention )
            dense_1= Dropout(0.1)(dense_1)
            dense_2=Dropout(0.1)(dense_1)
            
            output_layer = Dense(self.classes, activation='softmax')(dense_2)
            model = Model(inputs=[text_input, audio_input], outputs=output_layer)

            return model
if self.modality == "bimodal_":
                self.train_x_  = [self.train_x_text, self.train_x_audio]
                self.val_x_ = [self.val_x_text, self.val_x_audio]
                self.test_x_ = [self.test_x_text, self.test_x_audio]
                model = self.get_bimodal_model_()
                model.compile(optimizer=sgd,metrics=['acc',f1_m], loss='categorical_crossentropy', sample_weight_mode='temporal')
model.fit(self.train_x_ , self.train_y, epochs=self.epochs,batch_size=self.batch_size,sample_weight=self.train_mask,shuffle=True, 
validation_data=(self.val_x_, self.val_y, self.val_mask))