search

abudesai / timeVAE

TimeVAE implementation in keras/tensorflow
MIT License
121 stars 25 forks source link

about how to ues generate data for prediction task #10

Open blht-chen opened 7 months ago

blht-chen commented 7 months ago

Hi!author. Thank you for all your work. I plan to use the model you proposed for time series data augmentation. But I have a question to consult with you. My task is a multivariate prediction and univariate time series prediction task, which involves predicting the battery capacity based on collected current, voltage, and temperature. I would like to use your model to expand the current, voltage, and temperature. However, when using the expanded data for prediction, will the corresponding labels still be the same, or will the power also be input into your proposed model for expansion? Looking forward to your answer! Thank you so much!

abudesai commented 7 months ago

Hi, Do you have labels per time step, or does the entire series have one label? Are the labels continuous-valued, or are they categorical?

Best, Abu

On Sun, Apr 28, 2024 at 5:19 AM blht-chen @.***> wrote:

Hi!author. Thank you for all your work. I plan to use the model you proposed for time series data augmentation. But I have a question to consult with you. My task is a multivariate prediction and univariate time series prediction task, which involves predicting the battery capacity based on collected current, voltage, and temperature. I would like to use your model to expand the current, voltage, and temperature. However, when using the expanded data for prediction, will the corresponding labels still be the same, or will the power also be input into your proposed model for expansion? Looking forward to your answer! Thank you so much!

— Reply to this email directly, view it on GitHub https://github.com/abudesai/timeVAE/issues/10, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHXIAM3QQIEQHL6PEQVHSDLY7TSNFAVCNFSM6AAAAABG46CAFGVHI2DSMVQWIX3LMV43ASLTON2WKOZSGI3DONJTGUYTAOI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

blht-chen commented 7 months ago

thanks for you answer!i have label per time step,and the labels are continuous-valued.

abudesai commented 7 months ago

In that case, you can just pass the target series as part of the multivariate dataset. Synthetic data generation is an unsupervised method, so it would make no distinction between the features and targets in your dataset. It should work without requiring any modification to the dataset or the TimeVAE model.

On Tue, Apr 30, 2024 at 3:47 AM blht-chen @.***> wrote:

thanks for you answer!i have label per time step,and the labels are continuous-valued.

— Reply to this email directly, view it on GitHub https://github.com/abudesai/timeVAE/issues/10#issuecomment-2084970146, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHXIAMZJJWYNVYI5D7YYKC3Y75ZFNAVCNFSM6AAAAABG46CAFGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAOBUHE3TAMJUGY . You are receiving this because you commented.Message ID: @.***>