如何从TFrecord创建批量来训练网络中的张量流？

I have data saved to a tfrecord file. it has 1000 samples and 2 features (one is input the other output). the input is shape [1,20] and output [1,10]. they were both created from flattened numpy arrays. I am trying to create batches from them so i can use them to train my network but i am unable to figure out how. I have data saved to a tfrecord file. it has 10