{
  "id": 244303,
  "title": "What optimizer is working best for you ?",
  "url": "/competitions/seti-breakthrough-listen/discussion/244303",
  "author_name": "Mithil Salunkhe",
  "post_date": "2021-06-06T04:43:08.342000",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>For me the best optimizer currently is the following<br>\n<code>lr = (tf.keras.experimental.CosineDecayRestarts(0.03, 400)),\nopt = tf.keras.optimizers.SGD(lr)</code>. I have tried Adam but it does not work for me. Is anybody else using SGD ?</p>",
  "messages": [
    {
      "id": 1338016,
      "postDate": "2021-06-06T05:38:45.030Z",
      "content": "<p>Adam always work for me, SGD also works but usually it will take much longer training time for convergence.</p>",
      "rawMarkdown": "Adam always work for me, SGD also works but usually it will take much longer training time for convergence.",
      "votes": 3
    },
    {
      "id": 1337976,
      "postDate": "2021-06-06T04:43:08.343Z",
      "content": "<p>For me the best optimizer currently is the following<br>\n<code>lr = (tf.keras.experimental.CosineDecayRestarts(0.03, 400)),\nopt = tf.keras.optimizers.SGD(lr)</code>. I have tried Adam but it does not work for me. Is anybody else using SGD ?</p>",
      "rawMarkdown": "For me the best optimizer currently is the following\n`lr = (tf.keras.experimental.CosineDecayRestarts(0.03, 400)),\nopt = tf.keras.optimizers.SGD(lr)`. I have tried Adam but it does not work for me. Is anybody else using SGD ?",
      "votes": 1
    },
    {
      "id": 1338053,
      "postDate": "2021-06-06T06:12:44.400Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1338016,
      "author_name": "Chenglu",
      "author_url": "",
      "post_date": "2021-06-06T05:38:45.030000",
      "content": "<p>Adam always work for me, SGD also works but usually it will take much longer training time for convergence.</p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 1338053,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-06-06T06:12:44.400000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1338016": "Adam always work for me, SGD also works but usually it will take much longer training time for convergence.",
    "1337976": "For me the best optimizer currently is the following\n`lr = (tf.keras.experimental.CosineDecayRestarts(0.03, 400)),\nopt = tf.keras.optimizers.SGD(lr)`. I have tried Adam but it does not work for me. Is anybody else using SGD ?",
    "1338053": ""
  }
}