alibaba / sumn-universal-user-representation Goto Github PK

@m2408gj @JieGu

First of all, nice written code!
As I read the code, I saw parts that needed some modifications or would be good to be updated, so I suggest them.

1. Could you add some documents or README like Getting started or Quick start? Since there is no example data in code, it's quite hard to start over.
2. Continuing from question 1, can you post example data? If you look at the code for training with local data, it appears to read the local_data/universal_transformer_sample.txt file. Is it possible to upload this file? or some other.
3. Seems using PAI-tensorflow, such as tf.data.TableRecordDataset function which is not supported in plain TF. Could you decouple the dependency of PAI-tensorflow and change to plain TF runnable code?
4. Some codes are missing such as LocalFileDataLoader in loader, which is calling from train_locally.py.

Thanks a lot!