{
  "id": 437068,
  "title": "What Do Isolated Nodes Represent in a Computational Graph ?",
  "url": "/competitions/predict-ai-model-runtime/discussion/437068",
  "author_name": "AbaoJiang",
  "post_date": "2023-09-05T08:43:02.833000",
  "votes": 4,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi everyone,</p>\n<p>When exploring data with graph statistics, I find out that some of the graphs have <strong>isolated nodes</strong>, which don't connect to any other node in the graph. Acting as the <strong>tensor operation</strong>, what do <strong>isolated nodes</strong> represent in a computational graph?</p>\n<p>All <code>.npz</code> files with isolated nodes are listed as follows,</p>\n<pre><code>File \n-&gt; Isolated nodes: {, , , , , }\nFile \n-&gt; Isolated nodes: {, , , }\nFile \n-&gt; Isolated nodes: {}\nFile \n-&gt; Isolated nodes: {, , , , , , , , , , , , , , , }\nFile \n-&gt; Isolated nodes: {, , , , , , , , , , , , , , , }\nFile \n-&gt; Isolated nodes: {, , , , , }\nFile \n-&gt; Isolated nodes: {, , , }\nFile \n-&gt; Isolated nodes: {, , , , , , , }\nFile \n-&gt; Isolated nodes: {, , , , , , , , , , , , , , , }\nFile \n-&gt; Isolated nodes: {}\nFile \n-&gt; Isolated nodes: {, , , , , , , , , , , , , , , }\n</code></pre>\n<p>For implementation, please see <a href=\"https://www.kaggle.com/code/abaojiang/google-fast-or-slow-detailed-eda?scriptVersionId=141986607\" target=\"_blank\">Google - Fast or Slow? - Detailed EDA</a></p>\n<p>If there's any misunderstanding, please let me know. Thanks!</p>",
  "messages": [
    {
      "id": 2424403,
      "postDate": "2023-09-05T08:43:02.833Z",
      "content": "<p>Hi everyone,</p>\n<p>When exploring data with graph statistics, I find out that some of the graphs have <strong>isolated nodes</strong>, which don't connect to any other node in the graph. Acting as the <strong>tensor operation</strong>, what do <strong>isolated nodes</strong> represent in a computational graph?</p>\n<p>All <code>.npz</code> files with isolated nodes are listed as follows,</p>\n<pre><code>File \n-&gt; Isolated nodes: {, , , , , }\nFile \n-&gt; Isolated nodes: {, , , }\nFile \n-&gt; Isolated nodes: {}\nFile \n-&gt; Isolated nodes: {, , , , , , , , , , , , , , , }\nFile \n-&gt; Isolated nodes: {, , , , , , , , , , , , , , , }\nFile \n-&gt; Isolated nodes: {, , , , , }\nFile \n-&gt; Isolated nodes: {, , , }\nFile \n-&gt; Isolated nodes: {, , , , , , , }\nFile \n-&gt; Isolated nodes: {, , , , , , , , , , , , , , , }\nFile \n-&gt; Isolated nodes: {}\nFile \n-&gt; Isolated nodes: {, , , , , , , , , , , , , , , }\n</code></pre>\n<p>For implementation, please see <a href=\"https://www.kaggle.com/code/abaojiang/google-fast-or-slow-detailed-eda?scriptVersionId=141986607\" target=\"_blank\">Google - Fast or Slow? - Detailed EDA</a></p>\n<p>If there's any misunderstanding, please let me know. Thanks!</p>",
      "rawMarkdown": "Hi everyone,\n\nWhen exploring data with graph statistics, I find out that some of the graphs have **isolated nodes**, which don't connect to any other node in the graph. Acting as the **tensor operation**, what do **isolated nodes** represent in a computational graph?\n\nAll `.npz` files with isolated nodes are listed as follows,\n\n```python\nFile \"data/raw/npz_all/npz/layout/xla/default/train/mask_rcnn_batch_4_bf16_img1408.npz\"\n-> Isolated nodes: {51, 52, 1786, 1787, 1821, 1822}\nFile \"data/raw/npz_all/npz/layout/xla/default/train/mask_rcnn_resnet50.4x4.bf16.performance.npz\"\n-> Isolated nodes: {896, 909, 910, 895}\nFile \"data/raw/npz_all/npz/layout/xla/default/train/mlperf_transformer.npz\"\n-> Isolated nodes: {7885}\nFile \"data/raw/npz_all/npz/layout/xla/default/train/mask_rcnn_batch_16_bf16_img1024.npz\"\n-> Isolated nodes: {1760, 1788, 1789, 1933, 1934, 1904, 1905, 1875, 1876, 1846, 1847, 1817, 1818, 1724, 1725, 1759}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mlperf_maskrcnn_batch_4.npz\"\n-> Isolated nodes: {7584, 164, 165, 198, 199, 7563, 7564, 7597, 7598, 212, 213, 184, 185, 7611, 7612, 7583}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mask_rcnn_batch_4_bf16_img1408.npz\"\n-> Isolated nodes: {51, 52, 1786, 1787, 1821, 1822}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mask_rcnn_resnet50.4x4.bf16.performance.npz\"\n-> Isolated nodes: {896, 909, 910, 895}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mlperf_maskrcnn_batch_2.npz\"\n-> Isolated nodes: {160, 7232, 7211, 139, 140, 7212, 7231, 159}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mlperf_maskrcnn_1_shard_batch_4.npz\"\n-> Isolated nodes: {7584, 164, 165, 198, 199, 7563, 7564, 7597, 7598, 212, 213, 184, 185, 7611, 7612, 7583}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mlperf_transformer.npz\"\n-> Isolated nodes: {7885}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mask_rcnn_batch_16_bf16_img1024.npz\"\n-> Isolated nodes: {1760, 1788, 1789, 1933, 1934, 1904, 1905, 1875, 1876, 1846, 1847, 1817, 1818, 1724, 1725, 1759}\n```\n\nFor implementation, please see [Google - Fast or Slow? - Detailed EDA](https://www.kaggle.com/code/abaojiang/google-fast-or-slow-detailed-eda?scriptVersionId=141986607)\n\nIf there's any misunderstanding, please let me know. Thanks!",
      "votes": 3
    },
    {
      "id": 2425354,
      "postDate": "2023-09-05T20:24:19.773Z",
      "content": "<p>I'm looking at the first graph \"mask_rcnn_batch_4_bf16_img1408\". It looks like those nodes are unused inputs to the program. Their opcodes are 63, which maps to \"parameter\" (input) op: <a href=\"https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/hlo_opcode.h#L157C6-L157C15\" target=\"_blank\">https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/hlo_opcode.h#L157C6-L157C15</a>.</p>",
      "rawMarkdown": "I'm looking at the first graph \"mask_rcnn_batch_4_bf16_img1408\". It looks like those nodes are unused inputs to the program. Their opcodes are 63, which maps to \"parameter\" (input) op: https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/hlo_opcode.h#L157C6-L157C15.",
      "votes": 1,
      "replies": [
        {
          "id": 2425382,
          "postDate": "2023-09-05T21:01:24.737Z",
          "content": "<p>Isolation in this graph will rationalize you aim,not only that will real explain the exact meaning </p>",
          "rawMarkdown": "Isolation in this graph will rationalize you aim,not only that will real explain the exact meaning "
        },
        {
          "id": 2426694,
          "postDate": "2023-09-06T18:51:11.563Z",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/mangpophothilimthana\" target=\"_blank\">@mangpophothilimthana</a>,</p>\n<p>I missed it out. Thanks for your clarification.</p>",
          "rawMarkdown": "Hi @mangpophothilimthana,\n\nI missed it out. Thanks for your clarification."
        }
      ]
    },
    {
      "id": 2427077,
      "postDate": "2023-09-07T04:41:31.870Z",
      "rawMarkdown": "",
      "votes": -1,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 2425354,
      "author_name": "Mangpo Phothilimthana",
      "author_url": "",
      "post_date": "2023-09-05T20:24:19.773000",
      "content": "<p>I'm looking at the first graph \"mask_rcnn_batch_4_bf16_img1408\". It looks like those nodes are unused inputs to the program. Their opcodes are 63, which maps to \"parameter\" (input) op: <a href=\"https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/hlo_opcode.h#L157C6-L157C15\" target=\"_blank\">https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/hlo_opcode.h#L157C6-L157C15</a>.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 2425382,
          "author_name": "Abdulkadir Aliyu",
          "author_url": "",
          "post_date": "2023-09-05T21:01:24.737000",
          "content": "<p>Isolation in this graph will rationalize you aim,not only that will real explain the exact meaning </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 2426694,
          "author_name": "AbaoJiang",
          "author_url": "",
          "post_date": "2023-09-06T18:51:11.563000",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/mangpophothilimthana\" target=\"_blank\">@mangpophothilimthana</a>,</p>\n<p>I missed it out. Thanks for your clarification.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2427077,
      "author_name": "",
      "author_url": "",
      "post_date": "2023-09-07T04:41:31.870000",
      "content": "",
      "votes": -1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2424403": "Hi everyone,\n\nWhen exploring data with graph statistics, I find out that some of the graphs have **isolated nodes**, which don't connect to any other node in the graph. Acting as the **tensor operation**, what do **isolated nodes** represent in a computational graph?\n\nAll `.npz` files with isolated nodes are listed as follows,\n\n```python\nFile \"data/raw/npz_all/npz/layout/xla/default/train/mask_rcnn_batch_4_bf16_img1408.npz\"\n-> Isolated nodes: {51, 52, 1786, 1787, 1821, 1822}\nFile \"data/raw/npz_all/npz/layout/xla/default/train/mask_rcnn_resnet50.4x4.bf16.performance.npz\"\n-> Isolated nodes: {896, 909, 910, 895}\nFile \"data/raw/npz_all/npz/layout/xla/default/train/mlperf_transformer.npz\"\n-> Isolated nodes: {7885}\nFile \"data/raw/npz_all/npz/layout/xla/default/train/mask_rcnn_batch_16_bf16_img1024.npz\"\n-> Isolated nodes: {1760, 1788, 1789, 1933, 1934, 1904, 1905, 1875, 1876, 1846, 1847, 1817, 1818, 1724, 1725, 1759}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mlperf_maskrcnn_batch_4.npz\"\n-> Isolated nodes: {7584, 164, 165, 198, 199, 7563, 7564, 7597, 7598, 212, 213, 184, 185, 7611, 7612, 7583}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mask_rcnn_batch_4_bf16_img1408.npz\"\n-> Isolated nodes: {51, 52, 1786, 1787, 1821, 1822}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mask_rcnn_resnet50.4x4.bf16.performance.npz\"\n-> Isolated nodes: {896, 909, 910, 895}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mlperf_maskrcnn_batch_2.npz\"\n-> Isolated nodes: {160, 7232, 7211, 139, 140, 7212, 7231, 159}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mlperf_maskrcnn_1_shard_batch_4.npz\"\n-> Isolated nodes: {7584, 164, 165, 198, 199, 7563, 7564, 7597, 7598, 212, 213, 184, 185, 7611, 7612, 7583}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mlperf_transformer.npz\"\n-> Isolated nodes: {7885}\nFile \"data/raw/npz_all/npz/layout/xla/random/train/mask_rcnn_batch_16_bf16_img1024.npz\"\n-> Isolated nodes: {1760, 1788, 1789, 1933, 1934, 1904, 1905, 1875, 1876, 1846, 1847, 1817, 1818, 1724, 1725, 1759}\n```\n\nFor implementation, please see [Google - Fast or Slow? - Detailed EDA](https://www.kaggle.com/code/abaojiang/google-fast-or-slow-detailed-eda?scriptVersionId=141986607)\n\nIf there's any misunderstanding, please let me know. Thanks!",
    "2425354": "I'm looking at the first graph \"mask_rcnn_batch_4_bf16_img1408\". It looks like those nodes are unused inputs to the program. Their opcodes are 63, which maps to \"parameter\" (input) op: https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/hlo_opcode.h#L157C6-L157C15.",
    "2427077": ""
  }
}