{
  "id": 445225,
  "title": "Extracting additional features from protobuf",
  "url": "/competitions/predict-ai-model-runtime/discussion/445225",
  "author_name": "",
  "post_date": "2023-10-05T22:00:05.482259100Z",
  "votes": 14,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I'd like to experiment with additional features from protobuf.  For example, \"lhs_contracting_dims\", \"rhs_contracting_dims\", \"lhs_batch_dims\", \"rhs_batch_dims\" are features as part of the DotGeneral operator (ref: <a href=\"https://www.tensorflow.org/xla/operation_semantics#dotgeneral\" target=\"_blank\">https://www.tensorflow.org/xla/operation_semantics#dotgeneral</a>)</p>\n<p>It seems that the current feature extraction code (ref: <a href=\"https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/featurizers.h#L390\" target=\"_blank\">https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/featurizers.h#L390</a>) already have \"lhs_contracting_dims\", \"rhs_contracting_dims\", \"lhs_batch_dims\", \"rhs_batch_dims\" defined, but there are somehow not part of the 140 features of the node.</p>\n<p>Can you shed some light on how to easily change the feature extraction code to enable extraction of these additional features?</p>",
  "messages": [
    {
      "id": "2468851",
      "postDate": "10/05/2023 22:00:05",
      "content": "<p>I'd like to experiment with additional features from protobuf.  For example, \"lhs_contracting_dims\", \"rhs_contracting_dims\", \"lhs_batch_dims\", \"rhs_batch_dims\" are features as part of the DotGeneral operator (ref: <a href=\"https://www.tensorflow.org/xla/operation_semantics#dotgeneral\" target=\"_blank\">https://www.tensorflow.org/xla/operation_semantics#dotgeneral</a>)</p>\n<p>It seems that the current feature extraction code (ref: <a href=\"https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/featurizers.h#L390\" target=\"_blank\">https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/featurizers.h#L390</a>) already have \"lhs_contracting_dims\", \"rhs_contracting_dims\", \"lhs_batch_dims\", \"rhs_batch_dims\" defined, but there are somehow not part of the 140 features of the node.</p>\n<p>Can you shed some light on how to easily change the feature extraction code to enable extraction of these additional features?</p>",
      "rawMarkdown": "I'd like to experiment with additional features from protobuf.  For example, \"lhs_contracting_dims\", \"rhs_contracting_dims\", \"lhs_batch_dims\", \"rhs_batch_dims\" are features as part of the DotGeneral operator (ref: https://www.tensorflow.org/xla/operation_semantics#dotgeneral)\n\nIt seems that the current feature extraction code (ref: https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/featurizers.h#L390) already have \"lhs_contracting_dims\", \"rhs_contracting_dims\", \"lhs_batch_dims\", \"rhs_batch_dims\" defined, but there are somehow not part of the 140 features of the node.\n\nCan you shed some light on how to easily change the feature extraction code to enable extraction of these additional features?",
      "votes": null
    },
    {
      "id": "2472115",
      "postDate": "10/07/2023 02:46:12",
      "content": "<p>Very good finding. Adding those features may indeed help! Can you try to follow this instruction: <a href=\"https://github.com/google-research-datasets/tpu_graphs/tree/main#graph-feature-extraction\" target=\"_blank\">https://github.com/google-research-datasets/tpu_graphs/tree/main#graph-feature-extraction</a> ?</p>",
      "rawMarkdown": "Very good finding. Adding those features may indeed help! Can you try to follow this instruction: https://github.com/google-research-datasets/tpu_graphs/tree/main#graph-feature-extraction ?",
      "votes": null
    },
    {
      "id": "2486750",
      "postDate": "10/18/2023 06:39:50",
      "content": "<p>May I ask did you successfully extract new features? If so, could you share any tips about it?</p>",
      "rawMarkdown": "May I ask did you successfully extract new features? If so, could you share any tips about it?",
      "votes": null
    },
    {
      "id": "2492867",
      "postDate": "10/23/2023 01:23:02",
      "content": "<p>What was the reasoning behind engineering features but not including them in the numpy files? Did they not help?</p>",
      "rawMarkdown": "What was the reasoning behind engineering features but not including them in the numpy files? Did they not help?",
      "votes": null
    },
    {
      "id": "2496067",
      "postDate": "10/23/2023 18:04:30",
      "content": "<p>They didn't help the past, but I ran that experiment a very long time ago, so things might have changed.</p>",
      "rawMarkdown": "They didn't help the past, but I ran that experiment a very long time ago, so things might have changed.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2472115,
      "author_name": "mangpophothilimthana",
      "author_url": "",
      "post_date": "10/07/2023 02:46:12",
      "content": "<p>Very good finding. Adding those features may indeed help! Can you try to follow this instruction: <a href=\"https://github.com/google-research-datasets/tpu_graphs/tree/main#graph-feature-extraction\" target=\"_blank\">https://github.com/google-research-datasets/tpu_graphs/tree/main#graph-feature-extraction</a> ?</p>",
      "votes": null,
      "replies": [
        {
          "id": 2492867,
          "author_name": "alexanderliao",
          "author_url": "",
          "post_date": "10/23/2023 01:23:02",
          "content": "<p>What was the reasoning behind engineering features but not including them in the numpy files? Did they not help?</p>",
          "votes": null,
          "replies": [
            {
              "id": 2496067,
              "author_name": "mangpophothilimthana",
              "author_url": "",
              "post_date": "10/23/2023 18:04:30",
              "content": "<p>They didn't help the past, but I ran that experiment a very long time ago, so things might have changed.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2486750,
      "author_name": "tonyarobertson",
      "author_url": "",
      "post_date": "10/18/2023 06:39:50",
      "content": "<p>May I ask did you successfully extract new features? If so, could you share any tips about it?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2468851": "I'd like to experiment with additional features from protobuf.  For example, \"lhs_contracting_dims\", \"rhs_contracting_dims\", \"lhs_batch_dims\", \"rhs_batch_dims\" are features as part of the DotGeneral operator (ref: https://www.tensorflow.org/xla/operation_semantics#dotgeneral)\n\nIt seems that the current feature extraction code (ref: https://github.com/google-research-datasets/tpu_graphs/blob/main/tpu_graphs/process_data/xla/featurizers.h#L390) already have \"lhs_contracting_dims\", \"rhs_contracting_dims\", \"lhs_batch_dims\", \"rhs_batch_dims\" defined, but there are somehow not part of the 140 features of the node.\n\nCan you shed some light on how to easily change the feature extraction code to enable extraction of these additional features?",
    "2472115": "Very good finding. Adding those features may indeed help! Can you try to follow this instruction: https://github.com/google-research-datasets/tpu_graphs/tree/main#graph-feature-extraction ?",
    "2486750": "May I ask did you successfully extract new features? If so, could you share any tips about it?",
    "2492867": "What was the reasoning behind engineering features but not including them in the numpy files? Did they not help?",
    "2496067": "They didn't help the past, but I ran that experiment a very long time ago, so things might have changed."
  },
  "source": "meta"
}