解决ValueError: need at least one array to concatenate

ValueError: need at least one array to concatenate

引发原因：从空数组进行连接会引发此错误

报错提示为：

TFT_Dataset类报错相关部分源代码为：

for e in train[entity_column].unique():
          entity_group = data[data[entity_column]==e]
          
          data_time_steps = len(entity_group)

          if data_time_steps >= decoder_steps:
            x = entity_group[input_columns].values.astype(np.float32)
            inputs.append(np.stack([x[i:data_time_steps - (decoder_steps - 1) + i, :] for i in range(decoder_steps)], axis=1))  

            y = entity_group[[target_column]].values.astype(np.float32)
            outputs.append(np.stack([y[i:data_time_steps - (decoder_steps - 1) + i, :] for i in range(decoder_steps)], axis=1))

            e = entity_group[[entity_column]].values.astype(np.float32)
            entity.append(np.stack([e[i:data_time_steps - (decoder_steps - 1) + i, :] for i in range(decoder_steps)], axis=1))

            t = entity_group[[time_column]].values.astype(np.int64)
            time.append(np.stack([t[i:data_time_steps - (decoder_steps - 1) + i, :] for i in range(decoder_steps)], axis=1))
        
            
        self.inputs = np.concatenate(inputs, axis=0)
        self.outputs = np.concatenate(outputs, axis=0)[:, encoder_steps:, :]
        self.entity = np.concatenate(entity, axis=0)
        self.time = np.concatenate(time, axis=0)
        self.active_inputs = np.ones_like(outputs)

原因：当data为valid的时候，data_time_steps < decoder_steps，故返回的input为空

解决办法：添加当data_time_steps < decoder_steps情况时的处理

           else:
                 return None

具体如下

for e in train[entity_column].unique():
          entity_group = data[data[entity_column]==e]
          
          data_time_steps = len(entity_group)

          if data_time_steps >= decoder_steps:
            x = entity_group[input_columns].values.astype(np.float32)
            inputs.append(np.stack([x[i:data_time_steps - (decoder_steps - 1) + i, :] for i in range(decoder_steps)], axis=1))  
            

            y = entity_group[[target_column]].values.astype(np.float32)
            outputs.append(np.stack([y[i:data_time_steps - (decoder_steps - 1) + i, :] for i in range(decoder_steps)], axis=1))

            e = entity_group[[entity_column]].values.astype(np.float32)
            entity.append(np.stack([e[i:data_time_steps - (decoder_steps - 1) + i, :] for i in range(decoder_steps)], axis=1))

            t = entity_group[[time_column]].values.astype(np.int64)
            time.append(np.stack([t[i:data_time_steps - (decoder_steps - 1) + i, :] for i in range(decoder_steps)], axis=1))
        
           else:
                 return None
            
        self.inputs = np.concatenate(inputs, axis=0)
        self.outputs = np.concatenate(outputs, axis=0)[:, encoder_steps:, :]
        self.entity = np.concatenate(entity, axis=0)
        self.time = np.concatenate(time, axis=0)
        self.active_inputs = np.ones_like(outputs)

报错解决