https://github.com/nixtla logo
#neural-forecast
Title
# neural-forecast
t

thomas delaunait

04/23/2023, 6:18 PM
Hello everyone. Thank you for your amazing work!!! I am facing an error using azure function and NHits model. Locally it works, but once published in the cloud, I have this message : « you tried to log -1 which is not currently supported. Try a dict or a scalar/tensor. ». It seems to be due to tensorboard that tries to log file without success. Do you know if there is a way to set log_tensorboard=False (it seems to be possible with darts implementation). Thank you very much for your help and once again amazing work!
f

fede (nixtla) (they/them)

04/27/2023, 3:26 AM
hey @thomas delaunait! Thank you for letting us know about the issue. Do you happen to have more information about the environment you’re using? (python version, torch/lightning/neuralforecast versions would be great) :)
t

thomas delaunait

04/27/2023, 4:06 PM
Thank you @fede (nixtla) (they/them) we are running on python 3.10, torch 2.0.0 , ligthning 2.0.0 and neuralforecast 1.5.0. OS is Linux and we are using azure function (Microsoft serverless compute service). Thank you again for your kind help
A quick update, i have tested all the neural method (AutoMLP,NBeats,NHiTS…) and I am always facing the same error ( try to log-1…) I also downgraded lightning to 1.9.5. Same problem. @fede (nixtla) (they/them) do you have any ideas ? Thank you very much
f

fede (nixtla) (they/them)

05/17/2023, 8:46 PM
hey @thomas delaunait! Were you able to solve the problem?
t

thomas delaunait

05/18/2023, 12:42 PM
Hello @fede (nixtla) (they/them) unfortunately not. I have tried multiple combination of torch and lightning version…I think the issue is more related to the architecture we are using (azure function) and the ability to have write access and log file. We will continue investigating. I have not seen other people using azure function with neural forecast. I will keep you posted if we find a solution
9 Views