{
  "id": 312386,
  "title": "2 ways to approach this comp using CNN's",
  "url": "/competitions/kaggle-pog-series-s01e02/discussion/312386",
  "author_name": "",
  "post_date": "2022-03-11T18:15:49.744570200Z",
  "votes": 10,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I have 2 approaches in mind for this comp using CNN's.</p>\n<p>Approach 1: Convert audio to mel spectogram and treat it as an image classification problem. (Conv2D on input images)</p>\n<p>Approach 2: Work on the original signal without any preprocessing using Conv1D.</p>",
  "messages": [
    {
      "id": "1719409",
      "postDate": "03/11/2022 18:15:49",
      "content": "<p>I have 2 approaches in mind for this comp using CNN's.</p>\n<p>Approach 1: Convert audio to mel spectogram and treat it as an image classification problem. (Conv2D on input images)</p>\n<p>Approach 2: Work on the original signal without any preprocessing using Conv1D.</p>",
      "rawMarkdown": "I have 2 approaches in mind for this comp using CNN's.\n\nApproach 1: Convert audio to mel spectogram and treat it as an image classification problem. (Conv2D on input images)\n\nApproach 2: Work on the original signal without any preprocessing using Conv1D.",
      "votes": null
    },
    {
      "id": "1719483",
      "postDate": "03/11/2022 19:49:42",
      "content": "<p>Hey Harveen! Great to see you on Kaggle :) </p>\n<p>I was going to throw a Transformer model as a baseline but this was a good reminder for me to start small :)</p>",
      "rawMarkdown": "Hey Harveen! Great to see you on Kaggle :) \n\nI was going to throw a Transformer model as a baseline but this was a good reminder for me to start small :)",
      "votes": null
    },
    {
      "id": "1719763",
      "postDate": "03/12/2022 05:02:00",
      "content": "<p>Hey Sanyam :)</p>\n<p>Transformer as a baseline, people with DGX think differently 😃</p>",
      "rawMarkdown": "Hey Sanyam :)\n\nTransformer as a baseline, people with DGX think differently 😃",
      "votes": null
    },
    {
      "id": "1719942",
      "postDate": "03/12/2022 09:36:01",
      "content": "<p>yeah maybe starting with waveNet[uses conv1d] is a good option. </p>\n<ul>\n<li>If you[ <a href=\"https://www.kaggle.com/init27\" target=\"_blank\">@init27</a> ] are thinking of training a <code>Transformer model</code> from scratch then its ok, as the competition rules do not allow you to use pretrained model.</li>\n</ul>\n<blockquote>\n  <p>No pretrained models. Everything needs to be created from scratch.</p>\n</blockquote>",
      "rawMarkdown": "yeah maybe starting with waveNet[uses conv1d] is a good option. \n\n- If you[ @init27 ] are thinking of training a `Transformer model` from scratch then its ok, as the competition rules do not allow you to use pretrained model.\n\n> No pretrained models. Everything needs to be created from scratch.",
      "votes": null
    },
    {
      "id": "1723270",
      "postDate": "03/15/2022 09:17:18",
      "content": "<p>LOL <a href=\"https://www.kaggle.com/harveenchadha\" target=\"_blank\">@harveenchadha</a>. The DGX post was a joke, I have a much smaller box. 😄</p>\n<p>Although I would say I still won't do as well as you would even if I had access to more compute-I'm looking forward to learning from your NLP Knowledge in this comp 🙏</p>",
      "rawMarkdown": "LOL @harveenchadha. The DGX post was a joke, I have a much smaller box. 😄\n\nAlthough I would say I still won't do as well as you would even if I had access to more compute-I'm looking forward to learning from your NLP Knowledge in this comp 🙏",
      "votes": null
    },
    {
      "id": "1723643",
      "postDate": "03/15/2022 15:36:46",
      "content": "<p>I know nothing Grandmaster, but will try my best 😃 </p>",
      "rawMarkdown": "I know nothing Grandmaster, but will try my best 😃",
      "votes": null
    },
    {
      "id": "1727926",
      "postDate": "03/18/2022 12:53:28",
      "content": "<p>Baseline notebooks for both the approaches are public now :)</p>",
      "rawMarkdown": "Baseline notebooks for both the approaches are public now :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1719483,
      "author_name": "init27",
      "author_url": "",
      "post_date": "03/11/2022 19:49:42",
      "content": "<p>Hey Harveen! Great to see you on Kaggle :) </p>\n<p>I was going to throw a Transformer model as a baseline but this was a good reminder for me to start small :)</p>",
      "votes": null,
      "replies": [
        {
          "id": 1719763,
          "author_name": "harveenchadha",
          "author_url": "",
          "post_date": "03/12/2022 05:02:00",
          "content": "<p>Hey Sanyam :)</p>\n<p>Transformer as a baseline, people with DGX think differently 😃</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1719942,
          "author_name": "soumya9977",
          "author_url": "",
          "post_date": "03/12/2022 09:36:01",
          "content": "<p>yeah maybe starting with waveNet[uses conv1d] is a good option. </p>\n<ul>\n<li>If you[ <a href=\"https://www.kaggle.com/init27\" target=\"_blank\">@init27</a> ] are thinking of training a <code>Transformer model</code> from scratch then its ok, as the competition rules do not allow you to use pretrained model.</li>\n</ul>\n<blockquote>\n  <p>No pretrained models. Everything needs to be created from scratch.</p>\n</blockquote>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1723270,
          "author_name": "init27",
          "author_url": "",
          "post_date": "03/15/2022 09:17:18",
          "content": "<p>LOL <a href=\"https://www.kaggle.com/harveenchadha\" target=\"_blank\">@harveenchadha</a>. The DGX post was a joke, I have a much smaller box. 😄</p>\n<p>Although I would say I still won't do as well as you would even if I had access to more compute-I'm looking forward to learning from your NLP Knowledge in this comp 🙏</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1723643,
          "author_name": "harveenchadha",
          "author_url": "",
          "post_date": "03/15/2022 15:36:46",
          "content": "<p>I know nothing Grandmaster, but will try my best 😃 </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1727926,
      "author_name": "harveenchadha",
      "author_url": "",
      "post_date": "03/18/2022 12:53:28",
      "content": "<p>Baseline notebooks for both the approaches are public now :)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1719409": "I have 2 approaches in mind for this comp using CNN's.\n\nApproach 1: Convert audio to mel spectogram and treat it as an image classification problem. (Conv2D on input images)\n\nApproach 2: Work on the original signal without any preprocessing using Conv1D.",
    "1719483": "Hey Harveen! Great to see you on Kaggle :) \n\nI was going to throw a Transformer model as a baseline but this was a good reminder for me to start small :)",
    "1719763": "Hey Sanyam :)\n\nTransformer as a baseline, people with DGX think differently 😃",
    "1719942": "yeah maybe starting with waveNet[uses conv1d] is a good option. \n\n- If you[ @init27 ] are thinking of training a `Transformer model` from scratch then its ok, as the competition rules do not allow you to use pretrained model.\n\n> No pretrained models. Everything needs to be created from scratch.",
    "1723270": "LOL @harveenchadha. The DGX post was a joke, I have a much smaller box. 😄\n\nAlthough I would say I still won't do as well as you would even if I had access to more compute-I'm looking forward to learning from your NLP Knowledge in this comp 🙏",
    "1723643": "I know nothing Grandmaster, but will try my best 😃",
    "1727926": "Baseline notebooks for both the approaches are public now :)"
  },
  "source": "meta"
}