{
  "id": 77143,
  "title": "How to train a model from scratch",
  "url": "/competitions/humpback-whale-identification/discussion/77143",
  "author_name": "",
  "post_date": "2019-01-10T02:02:59.918645600Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I want to train a model from scratch, but how to decide the training strategy ,like epochs,learning rate etc, really confuses me.Any papers or posts will be greatly appreciated!</p>",
  "messages": [
    {
      "id": "453310",
      "postDate": "01/10/2019 02:02:59",
      "content": "<p>I want to train a model from scratch, but how to decide the training strategy ,like epochs,learning rate etc, really confuses me.Any papers or posts will be greatly appreciated!</p>",
      "rawMarkdown": "I want to train a model from scratch, but how to decide the training strategy ,like epochs,learning rate etc, really confuses me.Any papers or posts will be greatly appreciated!",
      "votes": null
    },
    {
      "id": "453315",
      "postDate": "01/10/2019 02:10:06",
      "content": "<p>You could listen the fastai videos and look at the kernels in this competition. This is how I started two years ago.</p>",
      "rawMarkdown": "You could listen the fastai videos and look at the kernels in this competition. This is how I started two years ago.",
      "votes": null
    },
    {
      "id": "453452",
      "postDate": "01/10/2019 08:06:58",
      "content": "<p>That always reamins a huge question for any data scientist. The best way to get around is to use Hyperparameter Optimization (HO). Take a look at these links, they discusses how to perform HO for sequential model of Keras.</p>\n\n<p><a href=\"https://towardsdatascience.com/hyperparameter-optimization-with-keras-b82e6364ca53\">Hyperparameter Optimization with Keras</a></p>\n\n<p><a href=\"https://machinelearningmastery.com/grid-search-hyperparameters-deep-learning-models-python-keras\">Grid Search Hyperparameters for Deep Learning Models in Python With Keras</a></p>\n\n<p>You can find similar articles if you are using R,</p>\n\n<p>Try searching for Grid/Random Search.</p>",
      "rawMarkdown": "That always reamins a huge question for any data scientist. The best way to get around is to use Hyperparameter Optimization (HO). Take a look at these links, they discusses how to perform HO for sequential model of Keras.\n\n[Hyperparameter Optimization with Keras][1]\n\n[Grid Search Hyperparameters for Deep Learning Models in Python With Keras][2]\n\nYou can find similar articles if you are using R,\n\nTry searching for Grid/Random Search.\n\n\n  [1]: https://towardsdatascience.com/hyperparameter-optimization-with-keras-b82e6364ca53\n  [2]: https://machinelearningmastery.com/grid-search-hyperparameters-deep-learning-models-python-keras",
      "votes": null
    },
    {
      "id": "453593",
      "postDate": "01/10/2019 12:53:55",
      "content": "<p>No, hyperparameter optimisation is the last thing you should do after:</p>\n\n<ul>\n<li>setuping your validation</li>\n<li>using data augmentation</li>\n<li>trying different model architectures</li>\n<li>tuning your loss function</li>\n<li>ensembling</li>\n</ul>",
      "rawMarkdown": "No, hyperparameter optimisation is the last thing you should do after:\n\n- setuping your validation\n- using data augmentation\n- trying different model architectures\n- tuning your loss function\n- ensembling",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 453315,
      "author_name": "chabir",
      "author_url": "",
      "post_date": "01/10/2019 02:10:06",
      "content": "<p>You could listen the fastai videos and look at the kernels in this competition. This is how I started two years ago.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 453452,
      "author_name": "shreyask92",
      "author_url": "",
      "post_date": "01/10/2019 08:06:58",
      "content": "<p>That always reamins a huge question for any data scientist. The best way to get around is to use Hyperparameter Optimization (HO). Take a look at these links, they discusses how to perform HO for sequential model of Keras.</p>\n\n<p><a href=\"https://towardsdatascience.com/hyperparameter-optimization-with-keras-b82e6364ca53\">Hyperparameter Optimization with Keras</a></p>\n\n<p><a href=\"https://machinelearningmastery.com/grid-search-hyperparameters-deep-learning-models-python-keras\">Grid Search Hyperparameters for Deep Learning Models in Python With Keras</a></p>\n\n<p>You can find similar articles if you are using R,</p>\n\n<p>Try searching for Grid/Random Search.</p>",
      "votes": null,
      "replies": [
        {
          "id": 453593,
          "author_name": "mratsim",
          "author_url": "",
          "post_date": "01/10/2019 12:53:55",
          "content": "<p>No, hyperparameter optimisation is the last thing you should do after:</p>\n\n<ul>\n<li>setuping your validation</li>\n<li>using data augmentation</li>\n<li>trying different model architectures</li>\n<li>tuning your loss function</li>\n<li>ensembling</li>\n</ul>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "453310": "I want to train a model from scratch, but how to decide the training strategy ,like epochs,learning rate etc, really confuses me.Any papers or posts will be greatly appreciated!",
    "453315": "You could listen the fastai videos and look at the kernels in this competition. This is how I started two years ago.",
    "453452": "That always reamins a huge question for any data scientist. The best way to get around is to use Hyperparameter Optimization (HO). Take a look at these links, they discusses how to perform HO for sequential model of Keras.\n\n[Hyperparameter Optimization with Keras][1]\n\n[Grid Search Hyperparameters for Deep Learning Models in Python With Keras][2]\n\nYou can find similar articles if you are using R,\n\nTry searching for Grid/Random Search.\n\n\n  [1]: https://towardsdatascience.com/hyperparameter-optimization-with-keras-b82e6364ca53\n  [2]: https://machinelearningmastery.com/grid-search-hyperparameters-deep-learning-models-python-keras",
    "453593": "No, hyperparameter optimisation is the last thing you should do after:\n\n- setuping your validation\n- using data augmentation\n- trying different model architectures\n- tuning your loss function\n- ensembling"
  },
  "source": "meta"
}