{
  "id": 176816,
  "title": "Just sharing some thoughts.Do this seem right?",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/176816",
  "author_name": "",
  "post_date": "2020-08-23T15:28:42.401477200Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>for each patient , using the ct scan (normalized)extract features using Conv and dense layers (say 10 features or less ) and then normalize the data for FCV  features in training_X data and then using the machine learning models to fit the image features + normalized fcv values in train_X to train_Y the not normalized FCV in training label. Then tune the parameters like a number of image features and other model parameters using cross-validation. Then to predict test_Y data again find the image features and normalize the FCV and predict using the trained model. I know I have missed some features. I would include them also in the model. <br>\nDo you think I am working in the right direction? And do you expect this plan to work?<br>\nSAY this works, even if I calculate the FCV I don't understand how do I calculate the confidence? Ideas on this are most welcomed!!!!</p>",
  "messages": [
    {
      "id": "982692",
      "postDate": "08/23/2020 15:28:42",
      "content": "<p>for each patient , using the ct scan (normalized)extract features using Conv and dense layers (say 10 features or less ) and then normalize the data for FCV  features in training_X data and then using the machine learning models to fit the image features + normalized fcv values in train_X to train_Y the not normalized FCV in training label. Then tune the parameters like a number of image features and other model parameters using cross-validation. Then to predict test_Y data again find the image features and normalize the FCV and predict using the trained model. I know I have missed some features. I would include them also in the model. <br>\nDo you think I am working in the right direction? And do you expect this plan to work?<br>\nSAY this works, even if I calculate the FCV I don't understand how do I calculate the confidence? Ideas on this are most welcomed!!!!</p>",
      "rawMarkdown": "for each patient , using the ct scan (normalized)extract features using Conv and dense layers (say 10 features or less ) and then normalize the data for FCV  features in training_X data and then using the machine learning models to fit the image features + normalized fcv values in train_X to train_Y the not normalized FCV in training label. Then tune the parameters like a number of image features and other model parameters using cross-validation. Then to predict test_Y data again find the image features and normalize the FCV and predict using the trained model. I know I have missed some features. I would include them also in the model. \nDo you think I am working in the right direction? And do you expect this plan to work?\nSAY this works, even if I calculate the FCV I don't understand how do I calculate the confidence? Ideas on this are most welcomed!!!!",
      "votes": null
    },
    {
      "id": "983550",
      "postDate": "08/24/2020 11:54:56",
      "content": "<p>Although not answer, I was just thinking about that,too.</p>\n<p>If train_y(target) set FVC,Model output is predicted FVC whitch  not  include confidence. </p>\n<p>if trained_model get confidence interval, have to change Models loss calculate with mezn and variance,I want to know what kind of  loss  method there is.</p>",
      "rawMarkdown": "Although not answer, I was just thinking about that,too.\n\nIf train_y(target) set FVC,Model output is predicted FVC whitch  not  include confidence. \n\nif trained_model get confidence interval, have to change Models loss calculate with mezn and variance,I want to know what kind of  loss  method there is.",
      "votes": null
    },
    {
      "id": "983590",
      "postDate": "08/24/2020 12:30:58",
      "content": "<p><a href=\"https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood\" target=\"_blank\">https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood</a></p>\n<p>With predicting each patients FVC,  can calculate laplace-log-likelihood  with actual FVC.<br>\nI dont understand what confidence value to set for output csv data.</p>",
      "rawMarkdown": "https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood\n\nWith predicting each patients FVC,  can calculate laplace-log-likelihood  with actual FVC.\nI dont understand what confidence value to set for output csv data.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 983550,
      "author_name": "yohehe",
      "author_url": "",
      "post_date": "08/24/2020 11:54:56",
      "content": "<p>Although not answer, I was just thinking about that,too.</p>\n<p>If train_y(target) set FVC,Model output is predicted FVC whitch  not  include confidence. </p>\n<p>if trained_model get confidence interval, have to change Models loss calculate with mezn and variance,I want to know what kind of  loss  method there is.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 983590,
      "author_name": "yohehe",
      "author_url": "",
      "post_date": "08/24/2020 12:30:58",
      "content": "<p><a href=\"https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood\" target=\"_blank\">https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood</a></p>\n<p>With predicting each patients FVC,  can calculate laplace-log-likelihood  with actual FVC.<br>\nI dont understand what confidence value to set for output csv data.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "982692": "for each patient , using the ct scan (normalized)extract features using Conv and dense layers (say 10 features or less ) and then normalize the data for FCV  features in training_X data and then using the machine learning models to fit the image features + normalized fcv values in train_X to train_Y the not normalized FCV in training label. Then tune the parameters like a number of image features and other model parameters using cross-validation. Then to predict test_Y data again find the image features and normalize the FCV and predict using the trained model. I know I have missed some features. I would include them also in the model. \nDo you think I am working in the right direction? And do you expect this plan to work?\nSAY this works, even if I calculate the FCV I don't understand how do I calculate the confidence? Ideas on this are most welcomed!!!!",
    "983550": "Although not answer, I was just thinking about that,too.\n\nIf train_y(target) set FVC,Model output is predicted FVC whitch  not  include confidence. \n\nif trained_model get confidence interval, have to change Models loss calculate with mezn and variance,I want to know what kind of  loss  method there is.",
    "983590": "https://www.kaggle.com/rohanrao/osic-understanding-laplace-log-likelihood\n\nWith predicting each patients FVC,  can calculate laplace-log-likelihood  with actual FVC.\nI dont understand what confidence value to set for output csv data."
  },
  "source": "meta"
}