How to add future covariates

[Weidong725]

My task is to carry out power load forecasting, the data is one data point per 15min (96 time points a day), I also have temperature data in the same format.After reading your source code, I found that the input have x_enc, x_mark_enc, but I want to use the historical load and temperature data of 480 points (5 days), combined with the forecast temperature of 96 points in the future, to predict the load of 96 points in the future.I would like to ask you how to add future variables to network.

[kwuking]

Thank you sincerely for your interest in our work and for posing such an insightful question. Our TimeMixer model is designed with the flexibility to seamlessly integrate future known features as covariates into the forecasting process. To achieve this, one would need to adjust the future_multi_mixing module to include x_mark_dec as part of the temporal embedding within the prediction framework. Your suggestion is highly valued, and in response, I am committed to releasing an updated version of our model that incorporates this functionality in the near future.

[Weidong725]

Thank you very much for your answer. Could you remind me after the update?

[kwuking]

Thank you very much for your answer. Could you remind me after the update?

Upon completion of the update, we will promptly notify you.

[Weidong725]

Besides, I have one more question. Taking my above task as an example, when forecasting power load, should we put the historical temperature data into x_enc or x_mark_enc?

1. If the historical temperature and load are put into the x_enc, whether to turn off the channel independently.
2. If the temperature and the daily type code are put into the x_mark_enc together, is it reasonable? which embedding method is more appropriate for this case?

[kwuking]

Thank you very much for your answer. Could you remind me after the update?

Hi, we have updated the code and released a new version to support future feature prediction. You can refer to the introduction in the README.

[kwuking]

Besides, I have one more question. Taking my above task as an example, when forecasting power load, should we put the historical temperature data into x_enc or x_mark_enc?

1. If the historical temperature and load are put into the x_enc, whether to turn off the channel independently.
2. If the temperature and the daily type code are put into the x_mark_enc together, is it reasonable? which embedding method is more appropriate for this case?

1. If you want the model to learn the relationship between the two, it is necessary to disable channel independence.

2. Obviously, both methods are feasible. If you put the temperature into x_enc, we will learn the variable relationship between it and the load. If you put it into x_mark_enc, we will use it as a future-known covariate to assist in prediction. Both methods are viable; it depends on our modeling approach and the actual results. You can try both, which might require slight modifications to the code to adapt to your scenario.

[Weidong725]

Thank you for your patient recovery. I'm ready to experiment with my task on the new version.

[qq234567890lyj]

你好，想问一下，复现这个代码需要改什么参数不，我这边复现出现问题了，想请教请教谢谢

[kwuking]

你好，想问一下，复现这个代码需要改什么参数不，我这边复现出现问题了，想请教请教谢谢

你好，看到了你的问题了，是否可以提供下具体的日志信息，以及执行的脚本和数据集的内容。

[wtt6668888]

在执行loss = criterion(outputs, batch_y)出现RuntimeError: The size of tensor a (96) must match the size of tensor b (144) at non-singleton dimension 1。就是在ETTh1数据集上执行的，还想请问一下这个模型可以利用多变量预测多变量吗？我的任务是一个时间点有ABC3个测量值，ABC之间是有联系的，之后预测ABC三个值，不知这个模型应当如何设置来预测呢？

[qq234567890lyj]

比如你输入（32，96，7）的数据，但是你中间做了其他转换，导致不是输入的96，你可以调试一下里面数据的变化，还有你可以改变那个feature为M就是多变量预测

…

---Original--- From: ***@***.***> Date: Wed, Sep 11, 2024 20:29 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [kwuking/TimeMixer] How to add future covariates (Issue #24) 在执行loss = criterion(outputs, batch_y)出现RuntimeError: The size of tensor a (96) must match the size of tensor b (144) at non-singleton dimension 1。就是在ETTh1数据集上执行的，还想请问一下这个模型可以利用多变量预测多变量吗？我的任务是一个时间点有ABC3个测量值，ABC之间是有联系的，之后预测ABC三个值，不知这个模型应当如何设置来预测呢？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>