{
  "id": 449137,
  "title": "Wrong data in layout/xla/random",
  "url": "/competitions/predict-ai-model-runtime/discussion/449137",
  "author_name": "no fit just luck",
  "post_date": "2023-10-23T09:30:41.532000",
  "votes": 11,
  "comment_count": 9,
  "views": 0,
  "content": "<p>We found that the data 'layout/xla/random/test/937ee0eb0d5d6151b7b8252933b5c1c9.npz' have a wrong size.</p>\n<pre><code>\nedgeruntime, shape is (1001,)\n</code></pre>",
  "messages": [
    {
      "id": 2493287,
      "postDate": "2023-10-23T09:30:41.533Z",
      "content": "<p>We found that the data 'layout/xla/random/test/937ee0eb0d5d6151b7b8252933b5c1c9.npz' have a wrong size.</p>\n<pre><code>\nedgeruntime, shape is (1001,)\n</code></pre>",
      "rawMarkdown": "We found that the data 'layout/xla/random/test/937ee0eb0d5d6151b7b8252933b5c1c9.npz' have a wrong size.\n```\n<numpy.lib.npyio.NpzFile object at 0x7efcefeecbe0>\n----------------\nedge_index, shape is (8694, 2)\nnode_feat, shape is (5279, 140)\nnode_opcode, shape is (5279,)\nnode_config_feat, shape is (1001, 161, 18)\nnode_config_ids, shape is (161,)\nnode_splits, shape is (1, 2)\nconfig_runtime, shape is (1001,)\n```",
      "votes": 10
    },
    {
      "id": 2496106,
      "postDate": "2023-10-23T18:20:01.443Z",
      "content": "<p>I just took a look. The data is correct here.</p>\n<p>there are 1001 sampled configurations (equal to runtimes). Configurable nodes are 161 and their indices are in <code>node_config_ids</code>.</p>\n<p>You are right, though: most others have 1000 sampled configurations. It looks like that there are some that have 1001 due to a numerical error. For example, on my macbookpro,</p>\n<pre><code>.arange(, (/)*, (/))\n</code></pre>\n<p>Gives a 1001-dimensional vector (however, mathematically, there should be only 1000 entries).</p>\n<p>Nonetheless: this is benign. The starter inference code scores all configurations. All of them are considered by the scoring script (when computing Kendal Tau correlation).</p>\n<p>However, thanks for bringing this up! I learned something new myself (about np.arange and numerical stability)</p>",
      "rawMarkdown": "I just took a look. The data is correct here.\n\nthere are 1001 sampled configurations (equal to runtimes). Configurable nodes are 161 and their indices are in `node_config_ids`.\n\nYou are right, though: most others have 1000 sampled configurations. It looks like that there are some that have 1001 due to a numerical error. For example, on my macbookpro,\n\n```\nnp.arange(0, (4983/1000)*1000, (4983/1000))\n```\n\nGives a 1001-dimensional vector (however, mathematically, there should be only 1000 entries).\n\nNonetheless: this is benign. The starter inference code scores all configurations. All of them are considered by the scoring script (when computing Kendal Tau correlation).\n\nHowever, thanks for bringing this up! I learned something new myself (about np.arange and numerical stability)",
      "votes": 3,
      "replies": [
        {
          "id": 2496117,
          "postDate": "2023-10-23T18:25:55.677Z",
          "content": "<p>Thank you so much!😀</p>",
          "rawMarkdown": "Thank you so much!😀"
        },
        {
          "id": 2496346,
          "postDate": "2023-10-24T01:51:03.300Z",
          "content": "<p>Thanks a lot!</p>",
          "rawMarkdown": "Thanks a lot!"
        }
      ]
    },
    {
      "id": 2493874,
      "postDate": "2023-10-23T16:06:14.457Z",
      "content": "<p>Hmm, which part is wrong? It looks okay to me.</p>",
      "rawMarkdown": "Hmm, which part is wrong? It looks okay to me.",
      "replies": [
        {
          "id": 2493901,
          "postDate": "2023-10-23T16:23:37.027Z",
          "content": "<p>Maybe something wrong with this <code>config_runtime, shape is (1001,)</code>? Others are (1000,).</p>",
          "rawMarkdown": "Maybe something wrong with this ``config_runtime, shape is (1001,)``? Others are (1000,).",
          "replies": [
            {
              "id": 2500848,
              "postDate": "2023-10-27T03:57:43.727Z",
              "rawMarkdown": "",
              "isDeleted": true
            }
          ]
        },
        {
          "id": 2493902,
          "postDate": "2023-10-23T16:23:43.900Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 2493833,
      "postDate": "2023-10-23T15:28:54.443Z",
      "content": "<p><a href=\"https://www.kaggle.com/mangpophothilimthana\" target=\"_blank\">@mangpophothilimthana</a> Maybe need a check?</p>",
      "rawMarkdown": "@mangpophothilimthana Maybe need a check?"
    },
    {
      "id": 2502339,
      "postDate": "2023-10-28T06:00:34.027Z",
      "content": "<p>Thanks for your reminder😀</p>",
      "rawMarkdown": "Thanks for your reminder😀"
    }
  ],
  "comments": [
    {
      "id": 2496106,
      "author_name": "Sami Abu-El-Haija",
      "author_url": "",
      "post_date": "2023-10-23T18:20:01.443000",
      "content": "<p>I just took a look. The data is correct here.</p>\n<p>there are 1001 sampled configurations (equal to runtimes). Configurable nodes are 161 and their indices are in <code>node_config_ids</code>.</p>\n<p>You are right, though: most others have 1000 sampled configurations. It looks like that there are some that have 1001 due to a numerical error. For example, on my macbookpro,</p>\n<pre><code>.arange(, (/)*, (/))\n</code></pre>\n<p>Gives a 1001-dimensional vector (however, mathematically, there should be only 1000 entries).</p>\n<p>Nonetheless: this is benign. The starter inference code scores all configurations. All of them are considered by the scoring script (when computing Kendal Tau correlation).</p>\n<p>However, thanks for bringing this up! I learned something new myself (about np.arange and numerical stability)</p>",
      "votes": 3,
      "replies": [
        {
          "id": 2496117,
          "author_name": "Zhecheng Li",
          "author_url": "",
          "post_date": "2023-10-23T18:25:55.677000",
          "content": "<p>Thank you so much!😀</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 2496346,
          "author_name": "no fit just luck",
          "author_url": "",
          "post_date": "2023-10-24T01:51:03.300000",
          "content": "<p>Thanks a lot!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2493874,
      "author_name": "Mangpo Phothilimthana",
      "author_url": "",
      "post_date": "2023-10-23T16:06:14.457000",
      "content": "<p>Hmm, which part is wrong? It looks okay to me.</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2493901,
          "author_name": "Zhecheng Li",
          "author_url": "",
          "post_date": "2023-10-23T16:23:37.027000",
          "content": "<p>Maybe something wrong with this <code>config_runtime, shape is (1001,)</code>? Others are (1000,).</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2500848,
              "author_name": "",
              "author_url": "",
              "post_date": "2023-10-27T03:57:43.727000",
              "content": "",
              "votes": 0,
              "replies": []
            }
          ]
        },
        {
          "id": 2493902,
          "author_name": "",
          "author_url": "",
          "post_date": "2023-10-23T16:23:43.900000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2493833,
      "author_name": "Zhecheng Li",
      "author_url": "",
      "post_date": "2023-10-23T15:28:54.443000",
      "content": "<p><a href=\"https://www.kaggle.com/mangpophothilimthana\" target=\"_blank\">@mangpophothilimthana</a> Maybe need a check?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2502339,
      "author_name": "kris",
      "author_url": "",
      "post_date": "2023-10-28T06:00:34.027000",
      "content": "<p>Thanks for your reminder😀</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2493287": "We found that the data 'layout/xla/random/test/937ee0eb0d5d6151b7b8252933b5c1c9.npz' have a wrong size.\n```\n<numpy.lib.npyio.NpzFile object at 0x7efcefeecbe0>\n----------------\nedge_index, shape is (8694, 2)\nnode_feat, shape is (5279, 140)\nnode_opcode, shape is (5279,)\nnode_config_feat, shape is (1001, 161, 18)\nnode_config_ids, shape is (161,)\nnode_splits, shape is (1, 2)\nconfig_runtime, shape is (1001,)\n```",
    "2496106": "I just took a look. The data is correct here.\n\nthere are 1001 sampled configurations (equal to runtimes). Configurable nodes are 161 and their indices are in `node_config_ids`.\n\nYou are right, though: most others have 1000 sampled configurations. It looks like that there are some that have 1001 due to a numerical error. For example, on my macbookpro,\n\n```\nnp.arange(0, (4983/1000)*1000, (4983/1000))\n```\n\nGives a 1001-dimensional vector (however, mathematically, there should be only 1000 entries).\n\nNonetheless: this is benign. The starter inference code scores all configurations. All of them are considered by the scoring script (when computing Kendal Tau correlation).\n\nHowever, thanks for bringing this up! I learned something new myself (about np.arange and numerical stability)",
    "2493874": "Hmm, which part is wrong? It looks okay to me.",
    "2493833": "@mangpophothilimthana Maybe need a check?",
    "2502339": "Thanks for your reminder😀"
  }
}