ByteType not supported, while IntegerType is too space-consuming

[wangjinjie722]

Hi there,

Under my training situation, the numerical value is small enough to use byte, which is better than int. But when I was trying save the byte type in dataframe to tfrecord, error came out "Caused by: java.lang.RuntimeException: Cannot convert field to unsupported data type ByteType". So why do we support ByteType? I know it's supported in sequence example, but why not support it in scalar type? Thanks!