{
  "id": 209407,
  "title": "How to choose model, optimizers, Loss functions ? ",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/209407",
  "author_name": "",
  "post_date": "2021-01-07T13:55:25.171455800Z",
  "votes": 3,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I tried lot getting best loss func/model/Optimizers. But now i started experimenting with less dataset. <br>\nI have tested<br>\nresnet 34 - 152<br>\nall effnets</p>\n<p>then some loss functions and optimizers </p>\n<p>But still having lot of optimzers and loss functions to test</p>\n<p>Do we have some specific structures ?<br>\nLike fist find best model on any loss/optimzer <br>\nthen find best loss on any optimzers or opposite ?</p>\n<p>How can we make this process less painful ?<br>\nWhat pros do ?</p>",
  "messages": [
    {
      "id": "1142615",
      "postDate": "01/07/2021 13:55:25",
      "content": "<p>I tried lot getting best loss func/model/Optimizers. But now i started experimenting with less dataset. <br>\nI have tested<br>\nresnet 34 - 152<br>\nall effnets</p>\n<p>then some loss functions and optimizers </p>\n<p>But still having lot of optimzers and loss functions to test</p>\n<p>Do we have some specific structures ?<br>\nLike fist find best model on any loss/optimzer <br>\nthen find best loss on any optimzers or opposite ?</p>\n<p>How can we make this process less painful ?<br>\nWhat pros do ?</p>",
      "rawMarkdown": "I tried lot getting best loss func/model/Optimizers. But now i started experimenting with less dataset. \nI have tested\nresnet 34 - 152\nall effnets\n\nthen some loss functions and optimizers \n\nBut still having lot of optimzers and loss functions to test\n\nDo we have some specific structures ?\nLike fist find best model on any loss/optimzer \nthen find best loss on any optimzers or opposite ?\n\nHow can we make this process less painful ?\nWhat pros do ?",
      "votes": null
    },
    {
      "id": "1142890",
      "postDate": "01/07/2021 16:41:57",
      "content": "<p>If you are using Tensorflow you could try out the keras hyperparameter tuner?</p>\n<p><a href=\"https://www.tensorflow.org/tutorials/keras/keras_tuner\" target=\"_blank\">https://www.tensorflow.org/tutorials/keras/keras_tuner</a></p>",
      "rawMarkdown": "If you are using Tensorflow you could try out the keras hyperparameter tuner?\n\nhttps://www.tensorflow.org/tutorials/keras/keras_tuner",
      "votes": null
    },
    {
      "id": "1143034",
      "postDate": "01/07/2021 18:16:41",
      "content": "<p>I am using pytorch. Thankyou for rply</p>",
      "rawMarkdown": "I am using pytorch. Thankyou for rply",
      "votes": null
    },
    {
      "id": "1143042",
      "postDate": "01/07/2021 18:20:33",
      "content": "<p>Optuna might be a good option then?</p>\n<p><a href=\"https://medium.com/pytorch/using-optuna-to-optimize-pytorch-hyperparameters-990607385e36\" target=\"_blank\">https://medium.com/pytorch/using-optuna-to-optimize-pytorch-hyperparameters-990607385e36</a></p>",
      "rawMarkdown": "Optuna might be a good option then?\n\nhttps://medium.com/pytorch/using-optuna-to-optimize-pytorch-hyperparameters-990607385e36",
      "votes": null
    },
    {
      "id": "1143074",
      "postDate": "01/07/2021 18:37:03",
      "content": "<p>Perfect<br>\nDo you recommend using Optuna on less data or full data ?</p>",
      "rawMarkdown": "Perfect\nDo you recommend using Optuna on less data or full data ?",
      "votes": null
    },
    {
      "id": "1143160",
      "postDate": "01/07/2021 19:17:11",
      "content": "<p>But but but…<br>\nWill optuna help me in selecting Loss or Optimers ?</p>",
      "rawMarkdown": "But but but...\nWill optuna help me in selecting Loss or Optimers ?",
      "votes": null
    },
    {
      "id": "1143226",
      "postDate": "01/07/2021 19:56:06",
      "content": "<p>Yeah you could definitely use it to tune optimizers.</p>\n<p>You might struggle with loss functions though because they are all measured on different scales. Hopefully somebody more experienced tuning loss can provide some insight?</p>",
      "rawMarkdown": "Yeah you could definitely use it to tune optimizers.\n\nYou might struggle with loss functions though because they are all measured on different scales. Hopefully somebody more experienced tuning loss can provide some insight?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1142890,
      "author_name": "brendanartley",
      "author_url": "",
      "post_date": "01/07/2021 16:41:57",
      "content": "<p>If you are using Tensorflow you could try out the keras hyperparameter tuner?</p>\n<p><a href=\"https://www.tensorflow.org/tutorials/keras/keras_tuner\" target=\"_blank\">https://www.tensorflow.org/tutorials/keras/keras_tuner</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 1143034,
          "author_name": "rajanlagah",
          "author_url": "",
          "post_date": "01/07/2021 18:16:41",
          "content": "<p>I am using pytorch. Thankyou for rply</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1143042,
          "author_name": "brendanartley",
          "author_url": "",
          "post_date": "01/07/2021 18:20:33",
          "content": "<p>Optuna might be a good option then?</p>\n<p><a href=\"https://medium.com/pytorch/using-optuna-to-optimize-pytorch-hyperparameters-990607385e36\" target=\"_blank\">https://medium.com/pytorch/using-optuna-to-optimize-pytorch-hyperparameters-990607385e36</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1143074,
          "author_name": "rajanlagah",
          "author_url": "",
          "post_date": "01/07/2021 18:37:03",
          "content": "<p>Perfect<br>\nDo you recommend using Optuna on less data or full data ?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1143160,
          "author_name": "rajanlagah",
          "author_url": "",
          "post_date": "01/07/2021 19:17:11",
          "content": "<p>But but but…<br>\nWill optuna help me in selecting Loss or Optimers ?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1143226,
          "author_name": "brendanartley",
          "author_url": "",
          "post_date": "01/07/2021 19:56:06",
          "content": "<p>Yeah you could definitely use it to tune optimizers.</p>\n<p>You might struggle with loss functions though because they are all measured on different scales. Hopefully somebody more experienced tuning loss can provide some insight?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1142615": "I tried lot getting best loss func/model/Optimizers. But now i started experimenting with less dataset. \nI have tested\nresnet 34 - 152\nall effnets\n\nthen some loss functions and optimizers \n\nBut still having lot of optimzers and loss functions to test\n\nDo we have some specific structures ?\nLike fist find best model on any loss/optimzer \nthen find best loss on any optimzers or opposite ?\n\nHow can we make this process less painful ?\nWhat pros do ?",
    "1142890": "If you are using Tensorflow you could try out the keras hyperparameter tuner?\n\nhttps://www.tensorflow.org/tutorials/keras/keras_tuner",
    "1143034": "I am using pytorch. Thankyou for rply",
    "1143042": "Optuna might be a good option then?\n\nhttps://medium.com/pytorch/using-optuna-to-optimize-pytorch-hyperparameters-990607385e36",
    "1143074": "Perfect\nDo you recommend using Optuna on less data or full data ?",
    "1143160": "But but but...\nWill optuna help me in selecting Loss or Optimers ?",
    "1143226": "Yeah you could definitely use it to tune optimizers.\n\nYou might struggle with loss functions though because they are all measured on different scales. Hopefully somebody more experienced tuning loss can provide some insight?"
  },
  "source": "meta"
}