Kael

Generally, there are two kinds of multi-step predictions: - single-shot model: it is controlled by [`y_dim`](https://github.com/kaelzhang/DA-RNN-in-Tensorflow-2-and-PyTorch/blob/master/da_rnn/torch/model.py#L248). For example, if you want to do 2-step no-shift predictions, you are required to...

Besides, IMO, time-series prediction is a very bad way to do quant trading and multi-step predictions will bring even worse results. I think a 96 time-step prediction will be worse...

This repo is just an experiment to implement the model in the paper, but the paper did not mention recursive prediction.

We could move discussion to #2 which is the exact same topic?

> At the same time, when making a forecast (in future), should I already know y-hist That is what DA-RNN does

Sorry that I supposed to have given it a completely refactor, using some build mechanism to make this module much simpler and much easier to extend, but I have been...

new major

@scaletimes 会更新的，因为基本上自己的小项目和公司的一些文档站点在用。 但是可能会在这个月晚点的时候开始，最近事情比较多。 下次更新的时候，会直接更新 [major](http://semver.org)（大版本） 版本到 1.0.0

@scaletimes 太棒了！之前写这个的时候还是很赶时间的完成任务。要做的东西还是不少的，建了不少 issue，应该会在3月底开始。