{
  "id": 436214,
  "title": "Layout Runtimes And Configurations",
  "url": "/competitions/predict-ai-model-runtime/discussion/436214",
  "author_name": "Mark Wijkhuizen",
  "post_date": "2023-09-01T11:48:31.320000",
  "votes": 7,
  "comment_count": 6,
  "views": 0,
  "content": "<p>When examining the layout data it is unclear to me how there can be 100040 runtimes.<br>\nThe layout runtimes are on a node level, so you would expect <code>C</code> different configuration for each node.<br>\nBut the number of runtimes is not divisible by 876.</p>\n<p>Could someone elaborate on the 100040 runtimes and further explain the <code>node_config_feat</code> in particular.</p>\n<p>Thanks in advance.</p>\n<pre><code>================ LAYOUT ================\nnode_feat             (, )\nnode_opcode           (,)\nedge_index            (, )\nnode_config_feat      (, , )\nnode_config_ids       (,)\nconfig_runtime        (,)\nnode_splits           (, )\n</code></pre>\n<p>Additional Information</p>\n<p><strong>node_config_ids</strong></p>\n<pre><code>, , , , , , , , , , , , , , , , , , , ,\n, , , , , , , , , , , , , , , , , , , \n</code></pre>",
  "messages": [
    {
      "id": 2418583,
      "postDate": "2023-09-01T11:48:31.320Z",
      "content": "<p>When examining the layout data it is unclear to me how there can be 100040 runtimes.<br>\nThe layout runtimes are on a node level, so you would expect <code>C</code> different configuration for each node.<br>\nBut the number of runtimes is not divisible by 876.</p>\n<p>Could someone elaborate on the 100040 runtimes and further explain the <code>node_config_feat</code> in particular.</p>\n<p>Thanks in advance.</p>\n<pre><code>================ LAYOUT ================\nnode_feat             (, )\nnode_opcode           (,)\nedge_index            (, )\nnode_config_feat      (, , )\nnode_config_ids       (,)\nconfig_runtime        (,)\nnode_splits           (, )\n</code></pre>\n<p>Additional Information</p>\n<p><strong>node_config_ids</strong></p>\n<pre><code>, , , , , , , , , , , , , , , , , , , ,\n, , , , , , , , , , , , , , , , , , , \n</code></pre>",
      "rawMarkdown": "When examining the layout data it is unclear to me how there can be 100040 runtimes.\nThe layout runtimes are on a node level, so you would expect `C` different configuration for each node.\nBut the number of runtimes is not divisible by 876.\n\nCould someone elaborate on the 100040 runtimes and further explain the `node_config_feat` in particular.\n\nThanks in advance.\n\n```\n================ LAYOUT ================\nnode_feat        \tshape (876, 140)\nnode_opcode      \tshape (876,)\nedge_index       \tshape (1366, 2)\nnode_config_feat \tshape (100040, 40, 18)\nnode_config_ids  \tshape (40,)\nconfig_runtime   \tshape (100040,)\nnode_splits      \tshape (1, 2)\n```\n\nAdditional Information\n\n**node_config_ids**\n\n```\n213, 223, 225, 231, 236, 246, 249, 253, 254, 258, 274, 282, 288, 293, 300, 321, 325, 328, 355, 374,\n399, 404, 409, 420, 424, 427, 452, 466, 491, 493, 499, 504, 622, 627, 674, 678, 730, 737, 839, 853\n```",
      "votes": 6
    },
    {
      "id": 2419140,
      "postDate": "2023-09-01T17:53:39.437Z",
      "content": "<p>The number of runtimes are not related to the number of nodes in the graph, but it depends on the number of configs.</p>\n<p>In this graph:<br>\nnumber of configs: c = 100040<br>\nnumber of nodes: n = 876<br>\nnumber of configurable nodes: nc = 40</p>\n<p>Essentially, the runtime of graph compiled with node_config_feat[i] is runtime [i], where 0 &lt;= i &lt; c.<br>\nIn this graph, 40 nodes are configurable. Therefore, each graph configuration (node_config_feat[i]) has 40 entries, where each entry is a node configuration (node_config_feat[i][j]), which is a feature vector of size 18.</p>\n<p>I hope this helps!</p>",
      "rawMarkdown": "The number of runtimes are not related to the number of nodes in the graph, but it depends on the number of configs.\n\nIn this graph:\nnumber of configs: c = 100040\nnumber of nodes: n = 876\nnumber of configurable nodes: nc = 40\n\nEssentially, the runtime of graph compiled with node_config_feat[i] is runtime [i], where 0 <= i < c.\nIn this graph, 40 nodes are configurable. Therefore, each graph configuration (node_config_feat[i]) has 40 entries, where each entry is a node configuration (node_config_feat[i][j]), which is a feature vector of size 18.\n\nI hope this helps!",
      "votes": 4,
      "replies": [
        {
          "id": 2419151,
          "postDate": "2023-09-01T18:04:08.883Z",
          "rawMarkdown": "",
          "isDeleted": true
        },
        {
          "id": 2419868,
          "postDate": "2023-09-02T08:47:43.173Z",
          "content": "<p>So out of the 876 nodes, 40 have a configurable layout.<br>\nThose 40 nodes are given in the <code>node_config_ids</code> as shown in the edited question.</p>\n<p>The <code>node_config_feat</code> thus gives 100040 different configurations for the 40 configurable nodes, each individual node configuration given as a 18 dimensional vector?</p>",
          "rawMarkdown": "So out of the 876 nodes, 40 have a configurable layout.\nThose 40 nodes are given in the `node_config_ids` as shown in the edited question.\n\nThe `node_config_feat` thus gives 100040 different configurations for the 40 configurable nodes, each individual node configuration given as a 18 dimensional vector?",
          "replies": [
            {
              "id": 2420537,
              "postDate": "2023-09-02T17:17:55.693Z",
              "content": "<p>That's right.</p>",
              "rawMarkdown": "That's right.",
              "votes": 1
            }
          ]
        }
      ]
    },
    {
      "id": 2419139,
      "postDate": "2023-09-01T17:52:55.213Z",
      "content": "<p>I think not all the nodes are configurable, for example out of 876 only 40 nodes are configurable and so we have node_config_feat as (100040, 40, 18).</p>",
      "rawMarkdown": "I think not all the nodes are configurable, for example out of 876 only 40 nodes are configurable and so we have node_config_feat as (100040, 40, 18)."
    },
    {
      "id": 2419911,
      "postDate": "2023-09-02T09:07:58.130Z",
      "content": "<p>Good info, thanks!</p>",
      "rawMarkdown": "Good info, thanks!"
    }
  ],
  "comments": [
    {
      "id": 2419140,
      "author_name": "Mangpo Phothilimthana",
      "author_url": "",
      "post_date": "2023-09-01T17:53:39.437000",
      "content": "<p>The number of runtimes are not related to the number of nodes in the graph, but it depends on the number of configs.</p>\n<p>In this graph:<br>\nnumber of configs: c = 100040<br>\nnumber of nodes: n = 876<br>\nnumber of configurable nodes: nc = 40</p>\n<p>Essentially, the runtime of graph compiled with node_config_feat[i] is runtime [i], where 0 &lt;= i &lt; c.<br>\nIn this graph, 40 nodes are configurable. Therefore, each graph configuration (node_config_feat[i]) has 40 entries, where each entry is a node configuration (node_config_feat[i][j]), which is a feature vector of size 18.</p>\n<p>I hope this helps!</p>",
      "votes": 4,
      "replies": [
        {
          "id": 2419151,
          "author_name": "",
          "author_url": "",
          "post_date": "2023-09-01T18:04:08.883000",
          "content": "",
          "votes": 0,
          "replies": []
        },
        {
          "id": 2419868,
          "author_name": "Mark Wijkhuizen",
          "author_url": "",
          "post_date": "2023-09-02T08:47:43.173000",
          "content": "<p>So out of the 876 nodes, 40 have a configurable layout.<br>\nThose 40 nodes are given in the <code>node_config_ids</code> as shown in the edited question.</p>\n<p>The <code>node_config_feat</code> thus gives 100040 different configurations for the 40 configurable nodes, each individual node configuration given as a 18 dimensional vector?</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2420537,
              "author_name": "Mangpo Phothilimthana",
              "author_url": "",
              "post_date": "2023-09-02T17:17:55.693000",
              "content": "<p>That's right.</p>",
              "votes": 1,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2419139,
      "author_name": "Pritam Sinha",
      "author_url": "",
      "post_date": "2023-09-01T17:52:55.213000",
      "content": "<p>I think not all the nodes are configurable, for example out of 876 only 40 nodes are configurable and so we have node_config_feat as (100040, 40, 18).</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2419911,
      "author_name": "ThomCat",
      "author_url": "",
      "post_date": "2023-09-02T09:07:58.130000",
      "content": "<p>Good info, thanks!</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2418583": "When examining the layout data it is unclear to me how there can be 100040 runtimes.\nThe layout runtimes are on a node level, so you would expect `C` different configuration for each node.\nBut the number of runtimes is not divisible by 876.\n\nCould someone elaborate on the 100040 runtimes and further explain the `node_config_feat` in particular.\n\nThanks in advance.\n\n```\n================ LAYOUT ================\nnode_feat        \tshape (876, 140)\nnode_opcode      \tshape (876,)\nedge_index       \tshape (1366, 2)\nnode_config_feat \tshape (100040, 40, 18)\nnode_config_ids  \tshape (40,)\nconfig_runtime   \tshape (100040,)\nnode_splits      \tshape (1, 2)\n```\n\nAdditional Information\n\n**node_config_ids**\n\n```\n213, 223, 225, 231, 236, 246, 249, 253, 254, 258, 274, 282, 288, 293, 300, 321, 325, 328, 355, 374,\n399, 404, 409, 420, 424, 427, 452, 466, 491, 493, 499, 504, 622, 627, 674, 678, 730, 737, 839, 853\n```",
    "2419140": "The number of runtimes are not related to the number of nodes in the graph, but it depends on the number of configs.\n\nIn this graph:\nnumber of configs: c = 100040\nnumber of nodes: n = 876\nnumber of configurable nodes: nc = 40\n\nEssentially, the runtime of graph compiled with node_config_feat[i] is runtime [i], where 0 <= i < c.\nIn this graph, 40 nodes are configurable. Therefore, each graph configuration (node_config_feat[i]) has 40 entries, where each entry is a node configuration (node_config_feat[i][j]), which is a feature vector of size 18.\n\nI hope this helps!",
    "2419139": "I think not all the nodes are configurable, for example out of 876 only 40 nodes are configurable and so we have node_config_feat as (100040, 40, 18).",
    "2419911": "Good info, thanks!"
  }
}