Tutorials on Colab crash

[Soffia-t]

Prerequisites

• [x ] Put an X between the brackets on this line if you have done all of the following:
  • Reproduced the problem in a new virtualenv with only neuralprophet installed, directly from github:

    git clone <copied link from github>
    cd neural_prophet
    pip install .

  • Checked the Answered Questions on the Github Disscussion board: https://github.com/ourownstory/neural_prophet/discussions
    If you have the same question but the Answer does not solve your issue, please continue the conversation there.
  • Checked that your issue isn't already filed: https://github.com/ourownstory/neural_prophet/issues
    If you have the same issue but there is a twist to your situation, please add an explanation there.
  • Considered whether your bug might actually be solveable by getting a question answered:
    • Please post a package use question
    • Please post a forecasting best practice question
    • Please post an idea or feedback

Describe the bug
When I try to run the tutorials (I tried several), for example (https://colab.research.google.com/github/ourownstory/neural_prophet/blob/main/docs/source/how-to-guides/application-examples/energy_hospital_load.ipynb) on Colab, they always crash the session (our of RAM message) when it tries to m.fit. What I have tried:
-Running a high ram Colab, same thing
-Running it on planetary computer, same thing
-Running on Noteable.io, same thing

I did not touch any of the code on the Tutorials.

P.S. It was happening on my notebooks using NeuralProphet, I thought it might be an error in my code, that's why I went to the tutorials and the error was happening there as well.

To Reproduce

Steps to reproduce the behavior:

1. Open notebook https://colab.research.google.com/github/ourownstory/neural_prophet/blob/main/docs/source/how-to-guides/application-examples/energy_hospital_load.ipynb on Colab
2. Run as is
3. It will crash the session when it tries to fit
4. Same thing happens when you use a High RAM machine or a GPU

Expected behavior

A fix to this so that we can run NeuralProphet on Colab

[leoniewgnr]

Hi @Soffia-t, thanks for raising this? Does it crash because it runs out of memory?

[Soffia-t]

That is what it says. That it ran out of memory.
But I have run the energy tutorial ( https://colab.research.google.com/github/ourownstory/neural_prophet/blob/main/docs/source/how-to-guides/application-examples/energy_hospital_load.ipynb) before ( on a normal Colab instance 4Gb) and it was fine. Now, it crashes even on the Pro+ version, 51 Gb of RAM.

[leoniewgnr]

I can confirm the problem, I'm currently working on improving the memory performance on NeuralProphet but didn't really find the problem. Do you have an idea where this could be in the fit function?
I suspect it to be in the time_dataset.py, but not super sure.

I already improved the stride function here. I'll merge it later and then we can check if it still runs out of memory, but I think so :(

[leoniewgnr]

This is the memory consumption of a normal fit, so I think we gradually add memory when loading the data and then for fitting it stays constant, which makes sense. I'll check the dataloading process if we can save memory somewhere.

[leoniewgnr]

@Soffia-t Do you have insights on how much RAM is used before it crashes?

[Soffia-t]

OK, So, I screen recorded the crash. I hope this will give you more info.
energy_data_example.ipynb - Colaboratory - Google Chrome 2023-08-16 09-58-16.zip

[leoniewgnr]

@Soffia-t, I made some improvements to memory issues, but it still crashes...
I honestly don't know why. Maybe someone can help on this?

[SimonWittner]

Hi,
I think this is an issue with the preinstalled packages from Colab. I tried it and I was able to fit the model after I uninstalled some of the packages before installing NeuralProphet:
!pip uninstall -y torch notebook notebook_shim tensorflow tensorflow-datasets prophet torchaudio torchdata torchtext torchvision
@leoniewgnr What do you think? Should we include this workaround in the tutorials?

[leoniewgnr]

resolved with #1418