Created by: pengwangucla

Tensorboard is a powerful tool for inspecting the training process ( checking parameters, intermedia results, accuracy and learning loss) This is especially useful when there is multiple loss and compare different model results. When training on server, we can view the process through webpages.

Currently, it is already supported for MXNet to combine tensorboard here. I think it could be also very useful for paddle, and it may not require much effort since both are dependent on Protobuf.

I wonder whether paddle is able to develop an api for it.

Thanks Peng Wang