{
  "id": 106076,
  "title": "Ensembling different models but ended with a LB score of .765",
  "url": "/competitions/aptos2019-blindness-detection/discussion/106076",
  "author_name": "",
  "post_date": "2019-08-28T04:13:22.704872800Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I ensembled the models by taking the average of their raw predictions. The were EfficientNets and a ResNet. I did preprocessing using ben's preprocessing along with data augmentations.\nI plan to  try training on different folds on the best scoring model and ensembling the folds. However, how should I ensemble so that I can increase my performance?  Thank you.</p>",
  "messages": [
    {
      "id": "609745",
      "postDate": "08/28/2019 04:13:22",
      "content": "<p>I ensembled the models by taking the average of their raw predictions. The were EfficientNets and a ResNet. I did preprocessing using ben's preprocessing along with data augmentations.\nI plan to  try training on different folds on the best scoring model and ensembling the folds. However, how should I ensemble so that I can increase my performance?  Thank you.</p>",
      "rawMarkdown": "I ensembled the models by taking the average of their raw predictions. The were EfficientNets and a ResNet. I did preprocessing using ben's preprocessing along with data augmentations.\nI plan to  try training on different folds on the best scoring model and ensembling the folds. However, how should I ensemble so that I can increase my performance?  Thank you.",
      "votes": null
    },
    {
      "id": "609776",
      "postDate": "08/28/2019 05:05:06",
      "content": "<p>what is your single model LB score? how many models have you ensembled?</p>",
      "rawMarkdown": "what is your single model LB score? how many models have you ensembled?",
      "votes": null
    },
    {
      "id": "609801",
      "postDate": "08/28/2019 05:50:37",
      "content": "<p>Try one(or all) of the ensembling approaches mentioned <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/104981#latest-606331\">here</a></p>",
      "rawMarkdown": "Try one(or all) of the ensembling approaches mentioned [here](https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/104981#latest-606331)",
      "votes": null
    },
    {
      "id": "610261",
      "postDate": "08/28/2019 15:50:04",
      "content": "<p>My single model LB score is .783 and I have ensembled 5 models.</p>",
      "rawMarkdown": "My single model LB score is .783 and I have ensembled 5 models.",
      "votes": null
    },
    {
      "id": "610366",
      "postDate": "08/28/2019 17:41:12",
      "content": "<p>You can try taking weighted average. I am using this:\nprediction = (prediction of model1 x model1 LB score + prediction of model2 x model2 LB score + prediction of model3 x model3 LB score + ...)/(model1 LB score + model2 LB score + model3 LB score ...)  </p>",
      "rawMarkdown": "You can try taking weighted average. I am using this:\nprediction = (prediction of model1 x model1 LB score + prediction of model2 x model2 LB score + prediction of model3 x model3 LB score + ...)/(model1 LB score + model2 LB score + model3 LB score ...)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 609776,
      "author_name": "aizhentech",
      "author_url": "",
      "post_date": "08/28/2019 05:05:06",
      "content": "<p>what is your single model LB score? how many models have you ensembled?</p>",
      "votes": null,
      "replies": [
        {
          "id": 610261,
          "author_name": "krsoju",
          "author_url": "",
          "post_date": "08/28/2019 15:50:04",
          "content": "<p>My single model LB score is .783 and I have ensembled 5 models.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 609801,
      "author_name": "bibek777",
      "author_url": "",
      "post_date": "08/28/2019 05:50:37",
      "content": "<p>Try one(or all) of the ensembling approaches mentioned <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/104981#latest-606331\">here</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 610366,
      "author_name": "tahsin",
      "author_url": "",
      "post_date": "08/28/2019 17:41:12",
      "content": "<p>You can try taking weighted average. I am using this:\nprediction = (prediction of model1 x model1 LB score + prediction of model2 x model2 LB score + prediction of model3 x model3 LB score + ...)/(model1 LB score + model2 LB score + model3 LB score ...)  </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "609745": "I ensembled the models by taking the average of their raw predictions. The were EfficientNets and a ResNet. I did preprocessing using ben's preprocessing along with data augmentations.\nI plan to  try training on different folds on the best scoring model and ensembling the folds. However, how should I ensemble so that I can increase my performance?  Thank you.",
    "609776": "what is your single model LB score? how many models have you ensembled?",
    "609801": "Try one(or all) of the ensembling approaches mentioned [here](https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/104981#latest-606331)",
    "610261": "My single model LB score is .783 and I have ensembled 5 models.",
    "610366": "You can try taking weighted average. I am using this:\nprediction = (prediction of model1 x model1 LB score + prediction of model2 x model2 LB score + prediction of model3 x model3 LB score + ...)/(model1 LB score + model2 LB score + model3 LB score ...)"
  },
  "source": "meta"
}