{
  "id": 403013,
  "title": "35th place solution (ensemble of public pre-trained models)",
  "url": "/competitions/icecube-neutrinos-in-deep-ice/discussion/403013",
  "author_name": "Emanuel Ruzak",
  "post_date": "2023-04-20T16:50:17.245000",
  "votes": 10,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I used the Graphnet baseline submission, and Robin Smits' ensemble of 3 LSTM<br>\nI created an ensemble model. It was trained with the outputs of the 4 models from batches 520-525, and the number of pulses in each event</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11784175%2Fa0feec28fba04174bcd4a30c2e18c184%2FDigaramEns2.drawio.png?generation=1682007580302032&amp;alt=media\" alt=\"\"></p>\n<p>The 3 outputs are (x,y,z) coordinates, the model was trained using the Mean Angular Error loss.</p>\n<p>This is the notebook used to train the ensemble</p>\n<p><a href=\"https://www.kaggle.com/code/emanuelruzak/ensemble2\" target=\"_blank\">https://www.kaggle.com/code/emanuelruzak/ensemble2</a></p>\n<p>This is the submission</p>\n<p><a href=\"https://www.kaggle.com/code/emanuelruzak/submissionfinal2\" target=\"_blank\">https://www.kaggle.com/code/emanuelruzak/submissionfinal2</a></p>\n<p>I also made a notebook to visualize the outputs of the LSTM models</p>\n<p><a href=\"https://www.kaggle.com/emanuelruzak/lstmviewicecube\" target=\"_blank\">https://www.kaggle.com/emanuelruzak/lstmviewicecube</a></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11784175%2F89b53b16eb15b5dcd0eeeaacf940ef0c%2FLSTMpred1.png?generation=1682009402450565&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": 2228619,
      "postDate": "2023-04-20T16:50:17.247Z",
      "content": "<p>I used the Graphnet baseline submission, and Robin Smits' ensemble of 3 LSTM<br>\nI created an ensemble model. It was trained with the outputs of the 4 models from batches 520-525, and the number of pulses in each event</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11784175%2Fa0feec28fba04174bcd4a30c2e18c184%2FDigaramEns2.drawio.png?generation=1682007580302032&amp;alt=media\" alt=\"\"></p>\n<p>The 3 outputs are (x,y,z) coordinates, the model was trained using the Mean Angular Error loss.</p>\n<p>This is the notebook used to train the ensemble</p>\n<p><a href=\"https://www.kaggle.com/code/emanuelruzak/ensemble2\" target=\"_blank\">https://www.kaggle.com/code/emanuelruzak/ensemble2</a></p>\n<p>This is the submission</p>\n<p><a href=\"https://www.kaggle.com/code/emanuelruzak/submissionfinal2\" target=\"_blank\">https://www.kaggle.com/code/emanuelruzak/submissionfinal2</a></p>\n<p>I also made a notebook to visualize the outputs of the LSTM models</p>\n<p><a href=\"https://www.kaggle.com/emanuelruzak/lstmviewicecube\" target=\"_blank\">https://www.kaggle.com/emanuelruzak/lstmviewicecube</a></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11784175%2F89b53b16eb15b5dcd0eeeaacf940ef0c%2FLSTMpred1.png?generation=1682009402450565&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "I used the Graphnet baseline submission, and Robin Smits' ensemble of 3 LSTM\nI created an ensemble model. It was trained with the outputs of the 4 models from batches 520-525, and the number of pulses in each event\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11784175%2Fa0feec28fba04174bcd4a30c2e18c184%2FDigaramEns2.drawio.png?generation=1682007580302032&alt=media)\n\n\nThe 3 outputs are (x,y,z) coordinates, the model was trained using the Mean Angular Error loss.\n\nThis is the notebook used to train the ensemble\n\nhttps://www.kaggle.com/code/emanuelruzak/ensemble2\n\nThis is the submission\n\nhttps://www.kaggle.com/code/emanuelruzak/submissionfinal2\n\nI also made a notebook to visualize the outputs of the LSTM models\n\nhttps://www.kaggle.com/emanuelruzak/lstmviewicecube\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11784175%2F89b53b16eb15b5dcd0eeeaacf940ef0c%2FLSTMpred1.png?generation=1682009402450565&alt=media)",
      "votes": 10
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2228619": "I used the Graphnet baseline submission, and Robin Smits' ensemble of 3 LSTM\nI created an ensemble model. It was trained with the outputs of the 4 models from batches 520-525, and the number of pulses in each event\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11784175%2Fa0feec28fba04174bcd4a30c2e18c184%2FDigaramEns2.drawio.png?generation=1682007580302032&alt=media)\n\n\nThe 3 outputs are (x,y,z) coordinates, the model was trained using the Mean Angular Error loss.\n\nThis is the notebook used to train the ensemble\n\nhttps://www.kaggle.com/code/emanuelruzak/ensemble2\n\nThis is the submission\n\nhttps://www.kaggle.com/code/emanuelruzak/submissionfinal2\n\nI also made a notebook to visualize the outputs of the LSTM models\n\nhttps://www.kaggle.com/emanuelruzak/lstmviewicecube\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11784175%2F89b53b16eb15b5dcd0eeeaacf940ef0c%2FLSTMpred1.png?generation=1682009402450565&alt=media)"
  }
}