nzl5116190 / Basisformer

This is the pytorch implementation of Basisformer in the Neurips paper: [BasisFormer: Attention-based Time Series Forecasting with Learnable and Interpretable Basis]
84 stars 8 forks source link

关于基和mark的来源? #11

Closed lilililwj closed 1 week ago

lilililwj commented 2 weeks ago

1.请问basis是如何得到的呢?

微信图片_20241113165236

这里的输入是什么? 2.这个mark指的是什么?它是如何得到的?

微信图片_20241113165421
nzl5116190 commented 2 weeks ago

English Reply:

Regarding question 1, as shown in the image below, we believe that the basis is a function of the timestamp. We take the normalized timestamp as input and pass it through a 4-layer Multi-Layer Perceptron (MLP) to obtain the basis functions.

Screenshot 2024-11-16 at 00 23 46

As for question 2, the "mark" refers to the timestamp, which is derived from the normalization process applied to the timestamps corresponding to each time point in the dataset. You can find the specific implementation in the data loader's code.


中文回复(Chinese version):

对于问题1,参考如下图片,我们认为basis是timestamp的一个函数。我们将归一化的timestamp作为输入,经过一个四层的多层感知机(MLP),从而得到基函数。

Screenshot 2024-11-16 at 00 23 46

对于问题2,mark指的是时间戳(timestamp),它是通过对数据集中每个时间点对应的时间戳进行归一化处理形成的,具体实现可以参考dataloader中的相关处理代码。

lilililwj commented 1 week ago

Thank you very much for your answer