{
  "id": 389541,
  "title": "Graphnet on CPU",
  "url": "/competitions/icecube-neutrinos-in-deep-ice/discussion/389541",
  "author_name": "",
  "post_date": "2023-02-22T05:24:04.714081300Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>While using graphnet observed that graphnet throws errors when trying to import :</p>\n<pre><code> graphnet.models.graph_builders  KNNGraphBuilder\n graphnet.models.task.reconstruction  (\n      AzimuthReconstructionWithKappa,\n      ZenithReconstruction,\n     )\n graphnet.training.loss_functions  VonMisesFisher2DLoss, CosineLoss\n graphnet.models.gnn.gnn  GNN\n graphnet.models.utils  calculate_xyzt_homophily\n graphnet.utilities.config  save_model_config\n</code></pre>\n<p>when system is on <code>CPU</code>, but works fine when using <code>GPU</code><br>\nIs it observed by anyone else? Is graphnet compatible only with <code>GPU</code> ?</p>",
  "messages": [
    {
      "id": "2154555",
      "postDate": "02/22/2023 05:24:04",
      "content": "<p>While using graphnet observed that graphnet throws errors when trying to import :</p>\n<pre><code> graphnet.models.graph_builders  KNNGraphBuilder\n graphnet.models.task.reconstruction  (\n      AzimuthReconstructionWithKappa,\n      ZenithReconstruction,\n     )\n graphnet.training.loss_functions  VonMisesFisher2DLoss, CosineLoss\n graphnet.models.gnn.gnn  GNN\n graphnet.models.utils  calculate_xyzt_homophily\n graphnet.utilities.config  save_model_config\n</code></pre>\n<p>when system is on <code>CPU</code>, but works fine when using <code>GPU</code><br>\nIs it observed by anyone else? Is graphnet compatible only with <code>GPU</code> ?</p>",
      "rawMarkdown": "While using graphnet observed that graphnet throws errors when trying to import :\n```python\nfrom graphnet.models.graph_builders import KNNGraphBuilder\nfrom graphnet.models.task.reconstruction import (\n      AzimuthReconstructionWithKappa,\n      ZenithReconstruction,\n     )\nfrom graphnet.training.loss_functions import VonMisesFisher2DLoss, CosineLoss\nfrom graphnet.models.gnn.gnn import GNN\nfrom graphnet.models.utils import calculate_xyzt_homophily\nfrom graphnet.utilities.config import save_model_config\n```\nwhen system is on `CPU`, but works fine when using `GPU`\nIs it observed by anyone else? Is graphnet compatible only with `GPU` ?",
      "votes": null
    },
    {
      "id": "2154902",
      "postDate": "02/22/2023 10:07:28",
      "content": "<p>You need to install CPU version of GraphNeT - example posted here features installation of the GPU version.</p>",
      "rawMarkdown": "You need to install CPU version of GraphNeT - example posted here features installation of the GPU version.",
      "votes": null
    },
    {
      "id": "2155156",
      "postDate": "02/22/2023 13:23:24",
      "content": "<p>Using CPU is extremely slow by the way, if you don't have enough GPU ressource on Kaggle consider using Colab instead, it will save you hours. Best</p>",
      "rawMarkdown": "Using CPU is extremely slow by the way, if you don't have enough GPU ressource on Kaggle consider using Colab instead, it will save you hours. Best",
      "votes": null
    },
    {
      "id": "2155294",
      "postDate": "02/22/2023 14:48:56",
      "content": "<p>Right, Thanks !</p>",
      "rawMarkdown": "Right, Thanks !",
      "votes": null
    },
    {
      "id": "2155305",
      "postDate": "02/22/2023 14:58:11",
      "content": "<p>I've made this notebook - use it instead of GPU-based one:<br>\n<a href=\"https://www.kaggle.com/atamazian/graphnet-and-dependencies-cpu\" target=\"_blank\">https://www.kaggle.com/atamazian/graphnet-and-dependencies-cpu</a></p>\n<p>And replace <code>pytorch</code> installation line with:<br>\n<code>!pip install /kaggle/working/software/dependencies/torch-1.11.0+cpu-cp37-cp37m-linux_x86_64.whl</code></p>\n<p>This could be useful when making SQLite databases from the training data.</p>",
      "rawMarkdown": "I've made this notebook - use it instead of GPU-based one:\nhttps://www.kaggle.com/atamazian/graphnet-and-dependencies-cpu\n\nAnd replace `pytorch` installation line with:\n`!pip install /kaggle/working/software/dependencies/torch-1.11.0+cpu-cp37-cp37m-linux_x86_64.whl`\n\nThis could be useful when making SQLite databases from the training data.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2154902,
      "author_name": "atamazian",
      "author_url": "",
      "post_date": "02/22/2023 10:07:28",
      "content": "<p>You need to install CPU version of GraphNeT - example posted here features installation of the GPU version.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2155294,
          "author_name": "himanshuwagh",
          "author_url": "",
          "post_date": "02/22/2023 14:48:56",
          "content": "<p>Right, Thanks !</p>",
          "votes": null,
          "replies": [
            {
              "id": 2155305,
              "author_name": "atamazian",
              "author_url": "",
              "post_date": "02/22/2023 14:58:11",
              "content": "<p>I've made this notebook - use it instead of GPU-based one:<br>\n<a href=\"https://www.kaggle.com/atamazian/graphnet-and-dependencies-cpu\" target=\"_blank\">https://www.kaggle.com/atamazian/graphnet-and-dependencies-cpu</a></p>\n<p>And replace <code>pytorch</code> installation line with:<br>\n<code>!pip install /kaggle/working/software/dependencies/torch-1.11.0+cpu-cp37-cp37m-linux_x86_64.whl</code></p>\n<p>This could be useful when making SQLite databases from the training data.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2155156,
      "author_name": "louisstefanuto",
      "author_url": "",
      "post_date": "02/22/2023 13:23:24",
      "content": "<p>Using CPU is extremely slow by the way, if you don't have enough GPU ressource on Kaggle consider using Colab instead, it will save you hours. Best</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2154555": "While using graphnet observed that graphnet throws errors when trying to import :\n```python\nfrom graphnet.models.graph_builders import KNNGraphBuilder\nfrom graphnet.models.task.reconstruction import (\n      AzimuthReconstructionWithKappa,\n      ZenithReconstruction,\n     )\nfrom graphnet.training.loss_functions import VonMisesFisher2DLoss, CosineLoss\nfrom graphnet.models.gnn.gnn import GNN\nfrom graphnet.models.utils import calculate_xyzt_homophily\nfrom graphnet.utilities.config import save_model_config\n```\nwhen system is on `CPU`, but works fine when using `GPU`\nIs it observed by anyone else? Is graphnet compatible only with `GPU` ?",
    "2154902": "You need to install CPU version of GraphNeT - example posted here features installation of the GPU version.",
    "2155156": "Using CPU is extremely slow by the way, if you don't have enough GPU ressource on Kaggle consider using Colab instead, it will save you hours. Best",
    "2155294": "Right, Thanks !",
    "2155305": "I've made this notebook - use it instead of GPU-based one:\nhttps://www.kaggle.com/atamazian/graphnet-and-dependencies-cpu\n\nAnd replace `pytorch` installation line with:\n`!pip install /kaggle/working/software/dependencies/torch-1.11.0+cpu-cp37-cp37m-linux_x86_64.whl`\n\nThis could be useful when making SQLite databases from the training data."
  },
  "source": "meta"
}