{
  "id": 21122,
  "title": "Is there any auto decreasing learning rate model",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/21122",
  "author_name": "",
  "post_date": "2016-05-21T10:54:48.893Z",
  "votes": null,
  "comment_count": 4,
  "views": 544,
  "content": "<p>Is there any variable learning rate model where learning rate starts with some high value and after each iteration if error increases then comes back a step and multiply learning rate to say 0.5 else keep same learning rate. </p>",
  "messages": [
    {
      "id": "120883",
      "postDate": "05/21/2016 10:54:48",
      "content": "<p>Is there any variable learning rate model where learning rate starts with some high value and after each iteration if error increases then comes back a step and multiply learning rate to say 0.5 else keep same learning rate. </p>",
      "rawMarkdown": "Is there any variable learning rate model where learning rate starts with some high value and after each iteration if error increases then comes back a step and multiply learning rate to say 0.5 else keep same learning rate.",
      "votes": null
    },
    {
      "id": "120885",
      "postDate": "05/21/2016 11:08:04",
      "content": "<p>I'm doing something very similar. If you're using Python and Keras, you can achieve that with a custom Callback class, based on the code they have for the ModelCheckpoint and EarlyStopping callbacks</p>",
      "rawMarkdown": "I'm doing something very similar. If you're using Python and Keras, you can achieve that with a custom Callback class, based on the code they have for the ModelCheckpoint and EarlyStopping callbacks",
      "votes": null
    },
    {
      "id": "120951",
      "postDate": "05/22/2016 06:32:39",
      "content": "<p>@ahs haez thanks for your return. Could you share with us the code ?</p>",
      "rawMarkdown": "ahs haez thanks for your return. Could you share with us the code ?",
      "votes": null
    },
    {
      "id": "120995",
      "postDate": "05/22/2016 14:32:47",
      "content": "<p>[quote=Mathurin Ach&#233;;120951]</p>\n\n<p>@ahs haez thanks for your return. Could you share with us the code ?</p>\n\n<p>[/quote]</p>\n\n<p>try this one: <a href=\"https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay\">https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay</a></p>",
      "rawMarkdown": "[quote=Mathurin Aché;120951]\r\n\r\n@ahs haez thanks for your return. Could you share with us the code ?\r\n\r\n[/quote]\r\n\r\ntry this one: https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay",
      "votes": null
    },
    {
      "id": "121003",
      "postDate": "05/22/2016 16:26:35",
      "content": "<p>[quote=ash hafez;120995]</p>\n\n<p>try this one: <a href=\"https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay\">https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay</a></p>\n\n<p>[/quote]\nthanks !</p>",
      "rawMarkdown": "[quote=ash hafez;120995]\r\n\r\ntry this one: https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay\r\n\r\n[/quote]\r\nthanks !",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 120885,
      "author_name": "ashhafez",
      "author_url": "",
      "post_date": "05/21/2016 11:08:04",
      "content": "<p>I'm doing something very similar. If you're using Python and Keras, you can achieve that with a custom Callback class, based on the code they have for the ModelCheckpoint and EarlyStopping callbacks</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 120951,
      "author_name": "mathurinache",
      "author_url": "",
      "post_date": "05/22/2016 06:32:39",
      "content": "<p>@ahs haez thanks for your return. Could you share with us the code ?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 120995,
      "author_name": "ashhafez",
      "author_url": "",
      "post_date": "05/22/2016 14:32:47",
      "content": "<p>[quote=Mathurin Ach&#233;;120951]</p>\n\n<p>@ahs haez thanks for your return. Could you share with us the code ?</p>\n\n<p>[/quote]</p>\n\n<p>try this one: <a href=\"https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay\">https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 121003,
      "author_name": "mathurinache",
      "author_url": "",
      "post_date": "05/22/2016 16:26:35",
      "content": "<p>[quote=ash hafez;120995]</p>\n\n<p>try this one: <a href=\"https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay\">https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay</a></p>\n\n<p>[/quote]\nthanks !</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "120883": "Is there any variable learning rate model where learning rate starts with some high value and after each iteration if error increases then comes back a step and multiply learning rate to say 0.5 else keep same learning rate.",
    "120885": "I'm doing something very similar. If you're using Python and Keras, you can achieve that with a custom Callback class, based on the code they have for the ModelCheckpoint and EarlyStopping callbacks",
    "120951": "ahs haez thanks for your return. Could you share with us the code ?",
    "120995": "[quote=Mathurin Aché;120951]\r\n\r\n@ahs haez thanks for your return. Could you share with us the code ?\r\n\r\n[/quote]\r\n\r\ntry this one: https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay",
    "121003": "[quote=ash hafez;120995]\r\n\r\ntry this one: https://www.kaggle.com/ashhafez/state-farm-distracted-driver-detection/keras-custom-learning-rate-decay\r\n\r\n[/quote]\r\nthanks !"
  },
  "source": "meta"
}