{
  "id": 191884,
  "title": "how to ensemble different submissions for private leaderboard",
  "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/191884",
  "author_name": "",
  "post_date": "2020-10-19T07:52:39.585674900Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello, I saw in public notebook, people just used several submission.csv to do the ensemble. But I think when it comes to private leaderboard, dong this would fail. For the code competition, we should load the pretrained model and do the inference. So if we ensemble several models, we should load all the models,  do the inferences, then ensemble the output from these models. But I am afraid the running time would exceed the maximum running hours. I am not sure whether I understand the rule correctly. Any idea is welcome. </p>",
  "messages": [
    {
      "id": "1053655",
      "postDate": "10/19/2020 07:52:39",
      "content": "<p>Hello, I saw in public notebook, people just used several submission.csv to do the ensemble. But I think when it comes to private leaderboard, dong this would fail. For the code competition, we should load the pretrained model and do the inference. So if we ensemble several models, we should load all the models,  do the inferences, then ensemble the output from these models. But I am afraid the running time would exceed the maximum running hours. I am not sure whether I understand the rule correctly. Any idea is welcome. </p>",
      "rawMarkdown": "Hello, I saw in public notebook, people just used several submission.csv to do the ensemble. But I think when it comes to private leaderboard, dong this would fail. For the code competition, we should load the pretrained model and do the inference. So if we ensemble several models, we should load all the models,  do the inferences, then ensemble the output from these models. But I am afraid the running time would exceed the maximum running hours. I am not sure whether I understand the rule correctly. Any idea is welcome.",
      "votes": null
    },
    {
      "id": "1053667",
      "postDate": "10/19/2020 08:10:08",
      "content": "<p>It is not really a code competition. They also shared the full test set (without the targets ofc). You could download it predict offline then upload your predictions. </p>\n<p><a href=\"https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/177662\" target=\"_blank\">Reduce submission time to ~10-15 minutes</a></p>\n<p><a href=\"https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/189043\" target=\"_blank\">Disappearing Inference Time Limitations</a></p>",
      "rawMarkdown": "It is not really a code competition. They also shared the full test set (without the targets ofc). You could download it predict offline then upload your predictions. \n\n[Reduce submission time to ~10-15 minutes](https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/177662)\n\n[Disappearing Inference Time Limitations](https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/189043)",
      "votes": null
    },
    {
      "id": "1055859",
      "postDate": "10/21/2020 07:01:08",
      "content": "<p>thank you!</p>",
      "rawMarkdown": "thank you!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1053667,
      "author_name": "gaborfodor",
      "author_url": "",
      "post_date": "10/19/2020 08:10:08",
      "content": "<p>It is not really a code competition. They also shared the full test set (without the targets ofc). You could download it predict offline then upload your predictions. </p>\n<p><a href=\"https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/177662\" target=\"_blank\">Reduce submission time to ~10-15 minutes</a></p>\n<p><a href=\"https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/189043\" target=\"_blank\">Disappearing Inference Time Limitations</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 1055859,
          "author_name": "shuozhang",
          "author_url": "",
          "post_date": "10/21/2020 07:01:08",
          "content": "<p>thank you!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1053655": "Hello, I saw in public notebook, people just used several submission.csv to do the ensemble. But I think when it comes to private leaderboard, dong this would fail. For the code competition, we should load the pretrained model and do the inference. So if we ensemble several models, we should load all the models,  do the inferences, then ensemble the output from these models. But I am afraid the running time would exceed the maximum running hours. I am not sure whether I understand the rule correctly. Any idea is welcome.",
    "1053667": "It is not really a code competition. They also shared the full test set (without the targets ofc). You could download it predict offline then upload your predictions. \n\n[Reduce submission time to ~10-15 minutes](https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/177662)\n\n[Disappearing Inference Time Limitations](https://www.kaggle.com/c/lyft-motion-prediction-autonomous-vehicles/discussion/189043)",
    "1055859": "thank you!"
  },
  "source": "meta"
}