About motivation and empirical analysis

[Irelandr]

Very good work! I have a question about the motivation for the experimental analysis: the evaluation of the optimal scheme involves data leakage, which seems bound to lead to significantly superior performance, so why not adopt a fairer comparison scheme? For example, directly measuring the distribution differences of time series.

[MogicianXD]

One motivation is that prior works including FSNet and OneNet just adopted the optimal scheme with data leakage. We would like to reveal the overestimation of their methods by comparing them with the actual performance under the practical scheme. We argue that we should not rely only on forecasting feedback on recent ground truth.