{ "id": 391454, "title": "Colab keeps crashing when importing graphnet modules", "url": "/competitions/icecube-neutrinos-in-deep-ice/discussion/391454", "author_name": "Michael Barrett", "post_date": "2023-03-01T14:14:22.531000", "votes": 1, "comment_count": 4, "views": 0, "content": "

Has anybody any experience importing GraphNeT into Colab? It took me a few days to work out how to set everything up and I've finally got:

\n

from graphnet.data.constants import FEATURES, TRUTH

\n

…working. But then when I tried the following two (in both/either order):

\n

from graphnet.models import StandardModel
\nfrom graphnet.models.gnn import DynEdge

\n

…I get \"Your session crashed for an unknown reason\" and the log doesn't appear to offer any useful information. I've already successfully run:

\n

!pip install -r requirements/torch_gpu.txt -e .[develop,torch]

\n

… so I assume I have the correct dependencies. I've also tried it with both GPU and CPU runtimes. Any suggestions?

", "messages": [ { "id": 2166673, "postDate": "2023-03-02T23:13:42.870Z", "content": "

The notebook suddenly closing might be due to some weird version errors. In which case I'd try the following (note that I haven't tried it, so it has some - non-trivial - chance of failing):

\n\n
!pip install --force-reinstall -r /path/to/graphnet/directory/requirements/torch_cpu.txt\n!pip install /path/to/graphnet/directory\n
\n\n
import torch\nprint(torch.__version__)\nimport torch_geometric\nprint(torch_geometric.__version__)\nimport graphnet\nprint(graphnet.__file__)\n
\n

For GPU, you might get some other problems, such as the CUDA version from graphnet not being the one installed in your environment. You can check the cuda version with import torch; print(torch.cuda.version) and install following instructions from the pytorch and pytorch-geometric websites, try to match the version from the graphnet requirements file (for example, if cuda is 11.3, install torch 1.11.0 for cuda 11.3).

\n

--force-reinstall might not be needed, I added just in case. You can add --quiet (after sorting all out) to silence pip.

\n

Cheers,
\nlts

\n

PS: It might be something related to memory too? So make sure you're not loading anything in the beginning, just get the imports right and only then start loading the files.

", "rawMarkdown": "The notebook suddenly closing might be due to some weird version errors. In which case I'd try the following (note that I haven't tried it, so it has some - non-trivial - chance of failing):\n\n- Restart the notebook completely (so we erase any bad imports)\n- Make sure these are the 2 first commands that the notebook run (so put it in a cell at the top):\n\n```\n!pip install --force-reinstall -r /path/to/graphnet/directory/requirements/torch_cpu.txt\n!pip install /path/to/graphnet/directory\n```\n\n- Test the imports:\n\n```\nimport torch\nprint(torch.__version__)\nimport torch_geometric\nprint(torch_geometric.__version__)\nimport graphnet\nprint(graphnet.__file__)\n```\n\nFor GPU, you might get some other problems, such as the CUDA version from graphnet not being the one installed in your environment. You can check the cuda version with `import torch; print(torch.cuda.version)` and install following instructions from the [pytorch](https://pytorch.org/get-started/previous-versions/) and [pytorch-geometric](https://pytorch-geometric.readthedocs.io/en/2.0.4/notes/installation.html) websites, try to match the version from the [graphnet requirements file](https://github.com/graphnet-team/graphnet/blob/main/requirements/torch_gpu.txt) (for example, if cuda is 11.3, install torch 1.11.0 for cuda 11.3).\n\n`--force-reinstall` might not be needed, I added just in case. You can add `--quiet` (after sorting all out) to silence pip.\n\nCheers,\nlts\n\nPS: It might be something related to memory too? So make sure you're not loading anything in the beginning, just get the imports right and only then start loading the files.", "votes": 1, "replies": [ { "id": 2167616, "postDate": "2023-03-03T15:57:23.700Z", "content": "

Thanks for your help.

\n

The problem related to numerous issues including (but not limited to): my lack of experience with Conda (or any other) environments and some Colab related idiosynchracies, as well as torch-geometric taking minutes to download - but then an hour to build and install.

\n

IIRC, that particular module (from graphnet.data.constants import FEATURES, TRUTH) stopped crashing when I made sure I ran this first:

\n

!pip install colorlog

", "rawMarkdown": "Thanks for your help.\n\nThe problem related to numerous issues including (but not limited to): my lack of experience with Conda (or any other) environments and some Colab related idiosynchracies, as well as torch-geometric taking minutes to download - but then an hour to build and install.\n\nIIRC, that particular module (from graphnet.data.constants import FEATURES, TRUTH) stopped crashing when I made sure I ran this first:\n\n`!pip install colorlog`", "replies": [ { "id": 2168054, "postDate": "2023-03-03T20:48:55.210Z", "content": "

Damn, I'm always sorry for people first coming into python package issues, it's always a f*cking mess haha In any case, good to know that you had your problem sorted, good luck in the competition !

", "rawMarkdown": "Damn, I'm always sorry for people first coming into python package issues, it's always a f*cking mess haha In any case, good to know that you had your problem sorted, good luck in the competition !", "votes": 1 } ] } ] }, { "id": 2164381, "postDate": "2023-03-01T14:14:22.530Z", "content": "

Has anybody any experience importing GraphNeT into Colab? It took me a few days to work out how to set everything up and I've finally got:

\n

from graphnet.data.constants import FEATURES, TRUTH

\n

…working. But then when I tried the following two (in both/either order):

\n

from graphnet.models import StandardModel
\nfrom graphnet.models.gnn import DynEdge

\n

…I get \"Your session crashed for an unknown reason\" and the log doesn't appear to offer any useful information. I've already successfully run:

\n

!pip install -r requirements/torch_gpu.txt -e .[develop,torch]

\n

… so I assume I have the correct dependencies. I've also tried it with both GPU and CPU runtimes. Any suggestions?

", "rawMarkdown": "Has anybody any experience importing GraphNeT into Colab? It took me a few days to work out how to set everything up and I've finally got: \n\n`from graphnet.data.constants import FEATURES, TRUTH`\n\n...working. But then when I tried the following two (in both/either order):\n\n`from graphnet.models import StandardModel`\n`from graphnet.models.gnn import DynEdge`\n\n...I get \"Your session crashed for an unknown reason\" and the log doesn't appear to offer any useful information. I've already successfully run:\n\n`!pip install -r requirements/torch_gpu.txt -e .[develop,torch]`\n\n... so I assume I have the correct dependencies. I've also tried it with both GPU and CPU runtimes. Any suggestions?", "votes": 1 }, { "id": 2164820, "postDate": "2023-03-01T20:13:22.753Z", "rawMarkdown": "", "votes": -1, "isDeleted": true } ], "comments": [ { "id": 2166673, "author_name": "lowerthansound", "author_url": "", "post_date": "2023-03-02T23:13:42.870000", "content": "

The notebook suddenly closing might be due to some weird version errors. In which case I'd try the following (note that I haven't tried it, so it has some - non-trivial - chance of failing):

\n\n
!pip install --force-reinstall -r /path/to/graphnet/directory/requirements/torch_cpu.txt\n!pip install /path/to/graphnet/directory\n
\n\n
import torch\nprint(torch.__version__)\nimport torch_geometric\nprint(torch_geometric.__version__)\nimport graphnet\nprint(graphnet.__file__)\n
\n

For GPU, you might get some other problems, such as the CUDA version from graphnet not being the one installed in your environment. You can check the cuda version with import torch; print(torch.cuda.version) and install following instructions from the pytorch and pytorch-geometric websites, try to match the version from the graphnet requirements file (for example, if cuda is 11.3, install torch 1.11.0 for cuda 11.3).

\n

--force-reinstall might not be needed, I added just in case. You can add --quiet (after sorting all out) to silence pip.

\n

Cheers,
\nlts

\n

PS: It might be something related to memory too? So make sure you're not loading anything in the beginning, just get the imports right and only then start loading the files.

", "votes": 1, "replies": [ { "id": 2167616, "author_name": "Michael Barrett", "author_url": "", "post_date": "2023-03-03T15:57:23.700000", "content": "

Thanks for your help.

\n

The problem related to numerous issues including (but not limited to): my lack of experience with Conda (or any other) environments and some Colab related idiosynchracies, as well as torch-geometric taking minutes to download - but then an hour to build and install.

\n

IIRC, that particular module (from graphnet.data.constants import FEATURES, TRUTH) stopped crashing when I made sure I ran this first:

\n

!pip install colorlog

", "votes": 0, "replies": [ { "id": 2168054, "author_name": "lowerthansound", "author_url": "", "post_date": "2023-03-03T20:48:55.210000", "content": "

Damn, I'm always sorry for people first coming into python package issues, it's always a f*cking mess haha In any case, good to know that you had your problem sorted, good luck in the competition !

", "votes": 1, "replies": [] } ] } ] }, { "id": 2164820, "author_name": "", "author_url": "", "post_date": "2023-03-01T20:13:22.753000", "content": "", "votes": -1, "replies": [] } ], "raw_markdown_by_id": { "2166673": "The notebook suddenly closing might be due to some weird version errors. In which case I'd try the following (note that I haven't tried it, so it has some - non-trivial - chance of failing):\n\n- Restart the notebook completely (so we erase any bad imports)\n- Make sure these are the 2 first commands that the notebook run (so put it in a cell at the top):\n\n```\n!pip install --force-reinstall -r /path/to/graphnet/directory/requirements/torch_cpu.txt\n!pip install /path/to/graphnet/directory\n```\n\n- Test the imports:\n\n```\nimport torch\nprint(torch.__version__)\nimport torch_geometric\nprint(torch_geometric.__version__)\nimport graphnet\nprint(graphnet.__file__)\n```\n\nFor GPU, you might get some other problems, such as the CUDA version from graphnet not being the one installed in your environment. You can check the cuda version with `import torch; print(torch.cuda.version)` and install following instructions from the [pytorch](https://pytorch.org/get-started/previous-versions/) and [pytorch-geometric](https://pytorch-geometric.readthedocs.io/en/2.0.4/notes/installation.html) websites, try to match the version from the [graphnet requirements file](https://github.com/graphnet-team/graphnet/blob/main/requirements/torch_gpu.txt) (for example, if cuda is 11.3, install torch 1.11.0 for cuda 11.3).\n\n`--force-reinstall` might not be needed, I added just in case. You can add `--quiet` (after sorting all out) to silence pip.\n\nCheers,\nlts\n\nPS: It might be something related to memory too? So make sure you're not loading anything in the beginning, just get the imports right and only then start loading the files.", "2164381": "Has anybody any experience importing GraphNeT into Colab? It took me a few days to work out how to set everything up and I've finally got: \n\n`from graphnet.data.constants import FEATURES, TRUTH`\n\n...working. But then when I tried the following two (in both/either order):\n\n`from graphnet.models import StandardModel`\n`from graphnet.models.gnn import DynEdge`\n\n...I get \"Your session crashed for an unknown reason\" and the log doesn't appear to offer any useful information. I've already successfully run:\n\n`!pip install -r requirements/torch_gpu.txt -e .[develop,torch]`\n\n... so I assume I have the correct dependencies. I've also tried it with both GPU and CPU runtimes. Any suggestions?", "2164820": "" } }