Closed alfredaita closed 1 month ago
Hi,
The example for StemGNN is indeed flawed, apologies.
from neuralforecast.models import StemGNN
Hope this helps, let me know.
Thank you for the quick response.Should it support these ?I was able to run the AirPassanger dataset on StemGNN without any exogenous list.The result were poor.
Sent from Yahoo Mail for iPhone
On Tuesday, July 30, 2024, 9:02 AM, Olivier Sprangers @.***> wrote:
Hi,
The example for StemGNN is indeed flawed, apologies.
Hope this helps, let me know.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
Thank you for the quick response.Should it support these ?I was able to run the AirPassanger dataset on StemGNN without any exogenous list.The result were poor.
No, it shouldn't support exogenous. And StemGNN is not the strongest model, and a bit hard to get good results from, so I can understand that the results are not that good. I'd advise you to start with TSMixerx
or MLPMultivariate
, if you want to use a multivariate with exogenous.
And a univariate model may perform better than a multivariate model, so just because the problem requires to forecast multiple (possibly correlated) timeseries, it doesn't mean a multivariate method is the best choice.
Thank you.Ive been modeling Bitcoin daily prices and have some success using LSTM with an exponential moving average and a simple moving average as a historical Exog lists .MSE ~ 3000. Better than most that I’ve tried but not good enough yet.
Sent from Yahoo Mail for iPhone
On Tuesday, July 30, 2024, 10:25 AM, Olivier Sprangers @.***> wrote:
Thank you for the quick response.Should it support these ?I was able to run the AirPassanger dataset on StemGNN without any exogenous list.The result were poor.
No, it shouldn't support exogenous. And StemGNN is not the strongest model, and a bit hard to get good results from, so I can understand that the results are not that good. I'd advise you to start with TSMixerxor MLPMultivariate, if you want to use a multivariate with exogenous.
And a univariate model may perform better than a multivariate model, so just because the problem requires to forecast multiple (possibly correlated) timeseries, it doesn't mean a multivariate method is the best choice.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
What happened + What you expected to happen
Following the example for StemGNN [copy paste or type] : 1) does not specify how to import StemGNN [ I did From neuralforecast.models import StemmGNN 2) when running the example : error StemGNN does not support futr_exog_list. 3) when I type StemGNN ? it shows that it does support it. 4) I expected it to run smoothly as the other models do.
NOTE: the summary table of all the models suggests StemGNN does not support futr.. [IE: They are not checked, ]
Versions / Dependencies
neuralforecast 1.7.3
I do not get this issue with other models from neuralforecast.models import
Reproduction script
The code is from neuralforecast StemGNN .
Issue Severity
High: It blocks me from completing my task.