XiaoHei

thanks very much! @aavbsouza , this seems great, I'll test this shared_dataset to see it's performance. Btw, I use plasma to store my whole data and read the whole data...

see it in the trainer init() func