{
  "id": 394320,
  "title": "Is that we can't ensemble too many models? And some question about the submodel class we ensemble",
  "url": "/competitions/asl-signs/discussion/394320",
  "author_name": "",
  "post_date": "2023-03-13T04:32:25.268945400Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I see that in the <code>evaluation</code> page, it says <code>Your model must also perform inference with less than 100 milliseconds of latency per video on average and use less than 40 MB of storage space</code>. Does it mean we can't ensemble too many models in case it will be out of 40MB? <br>\nand one more question, we know that the model in submission.zip should be a <strong>tflite model</strong> class, but if we ensemble <em>model_a</em>, <em>model_b</em>, balabala in a tflite model, does it require <em>model_a</em>, <em>model_b</em> should also be tflite or they can be common tf.keras.Model class?</p>",
  "messages": [
    {
      "id": "2179307",
      "postDate": "03/13/2023 04:32:25",
      "content": "<p>I see that in the <code>evaluation</code> page, it says <code>Your model must also perform inference with less than 100 milliseconds of latency per video on average and use less than 40 MB of storage space</code>. Does it mean we can't ensemble too many models in case it will be out of 40MB? <br>\nand one more question, we know that the model in submission.zip should be a <strong>tflite model</strong> class, but if we ensemble <em>model_a</em>, <em>model_b</em>, balabala in a tflite model, does it require <em>model_a</em>, <em>model_b</em> should also be tflite or they can be common tf.keras.Model class?</p>",
      "rawMarkdown": "I see that in the `evaluation` page, it says `Your model must also perform inference with less than 100 milliseconds of latency per video on average and use less than 40 MB of storage space`. Does it mean we can't ensemble too many models in case it will be out of 40MB? \nand one more question, we know that the model in submission.zip should be a **tflite model** class, but if we ensemble *model_a*, *model_b*, balabala in a tflite model, does it require *model_a*, *model_b* should also be tflite or they can be common tf.keras.Model class?",
      "votes": null
    },
    {
      "id": "2180741",
      "postDate": "03/14/2023 04:13:25",
      "content": "<p>here is good example how to ensemble <a href=\"https://www.kaggle.com/code/dschettler8845/gislr-how-to-ensemble\" target=\"_blank\">https://www.kaggle.com/code/dschettler8845/gislr-how-to-ensemble</a></p>",
      "rawMarkdown": "here is good example how to ensemble https://www.kaggle.com/code/dschettler8845/gislr-how-to-ensemble",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2180741,
      "author_name": "programmaticart",
      "author_url": "",
      "post_date": "03/14/2023 04:13:25",
      "content": "<p>here is good example how to ensemble <a href=\"https://www.kaggle.com/code/dschettler8845/gislr-how-to-ensemble\" target=\"_blank\">https://www.kaggle.com/code/dschettler8845/gislr-how-to-ensemble</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2179307": "I see that in the `evaluation` page, it says `Your model must also perform inference with less than 100 milliseconds of latency per video on average and use less than 40 MB of storage space`. Does it mean we can't ensemble too many models in case it will be out of 40MB? \nand one more question, we know that the model in submission.zip should be a **tflite model** class, but if we ensemble *model_a*, *model_b*, balabala in a tflite model, does it require *model_a*, *model_b* should also be tflite or they can be common tf.keras.Model class?",
    "2180741": "here is good example how to ensemble https://www.kaggle.com/code/dschettler8845/gislr-how-to-ensemble"
  },
  "source": "meta"
}