{
  "id": 43733,
  "title": "Raspberry Pi Leaderboard? ",
  "url": "/competitions/tensorflow-speech-recognition-challenge/discussion/43733",
  "author_name": "",
  "post_date": "2017-11-18T16:50:38.477101800Z",
  "votes": 9,
  "comment_count": 5,
  "views": 0,
  "content": "<p>How would special models be recognized? </p>\n\n<p>Assuming a model met all qualifications for the special prize, how would the person who made it know if its accuracy is \"good enough\" compared to other \"special\" models?</p>\n\n<p>Since it'll have a low score on the Public Leaderboard relative to other \"normal\" submissions. </p>",
  "messages": [
    {
      "id": "245497",
      "postDate": "11/18/2017 16:50:38",
      "content": "<p>How would special models be recognized? </p>\n\n<p>Assuming a model met all qualifications for the special prize, how would the person who made it know if its accuracy is \"good enough\" compared to other \"special\" models?</p>\n\n<p>Since it'll have a low score on the Public Leaderboard relative to other \"normal\" submissions. </p>",
      "rawMarkdown": "How would special models be recognized? \n\nAssuming a model met all qualifications for the special prize, how would the person who made it know if its accuracy is \"good enough\" compared to other \"special\" models?\n\nSince it'll have a low score on the Public Leaderboard relative to other \"normal\" submissions.",
      "votes": null
    },
    {
      "id": "245524",
      "postDate": "11/18/2017 18:25:20",
      "content": "<p>It would be nice, but it would require a lot of setup on Kaggle's end to automatically run code submissions on a pi, and I suspect a lot of the best pi-sized submissions will come later in the competition when people trim down their regular models.</p>",
      "rawMarkdown": "It would be nice, but it would require a lot of setup on Kaggle's end to automatically run code submissions on a pi, and I suspect a lot of the best pi-sized submissions will come later in the competition when people trim down their regular models.",
      "votes": null
    },
    {
      "id": "245760",
      "postDate": "11/19/2017 15:50:34",
      "content": "<p>Requiring a lot of setup on Kaggle's end is the problem, though I'm interested to see what they'll come up with if they choose not to make another leaderboard or solution checker. </p>",
      "rawMarkdown": "Requiring a lot of setup on Kaggle's end is the problem, though I'm interested to see what they'll come up with if they choose not to make another leaderboard or solution checker.",
      "votes": null
    },
    {
      "id": "246772",
      "postDate": "11/21/2017 18:53:14",
      "content": "<p>The main question I have is whether the raspberry pi solution must be selected as one of the final two submissions, and whether it has to be our highest scoring submission of the two submissions selected. </p>",
      "rawMarkdown": "The main question I have is whether the raspberry pi solution must be selected as one of the final two submissions, and whether it has to be our highest scoring submission of the two submissions selected.",
      "votes": null
    },
    {
      "id": "246776",
      "postDate": "11/21/2017 19:00:02",
      "content": "<p>Devin,</p>\n\n<p>Submissions for the Raspberry Pi Solution will be separately submitted and do not need to be one of your final two submissions.</p>",
      "rawMarkdown": "Devin,\n\nSubmissions for the Raspberry Pi Solution will be separately submitted and do not need to be one of your final two submissions.",
      "votes": null
    },
    {
      "id": "246860",
      "postDate": "11/21/2017 21:41:31",
      "content": "<p>I'm working on a solution compatible with the RPI3 description. My current model has 3,1MB, inference time &lt; 100ms and is achieving 76% accuracy. The model was trained for less than an hour. It's overfitting the training data, so I'll try to augment the data. Let's see what happens.</p>",
      "rawMarkdown": "I'm working on a solution compatible with the RPI3 description. My current model has 3,1MB, inference time &lt; 100ms and is achieving 76% accuracy. The model was trained for less than an hour. It's overfitting the training data, so I'll try to augment the data. Let's see what happens.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 245524,
      "author_name": "happycube",
      "author_url": "",
      "post_date": "11/18/2017 18:25:20",
      "content": "<p>It would be nice, but it would require a lot of setup on Kaggle's end to automatically run code submissions on a pi, and I suspect a lot of the best pi-sized submissions will come later in the competition when people trim down their regular models.</p>",
      "votes": null,
      "replies": [
        {
          "id": 245760,
          "author_name": "mmi333",
          "author_url": "",
          "post_date": "11/19/2017 15:50:34",
          "content": "<p>Requiring a lot of setup on Kaggle's end is the problem, though I'm interested to see what they'll come up with if they choose not to make another leaderboard or solution checker. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 246772,
      "author_name": "devinanzelmo",
      "author_url": "",
      "post_date": "11/21/2017 18:53:14",
      "content": "<p>The main question I have is whether the raspberry pi solution must be selected as one of the final two submissions, and whether it has to be our highest scoring submission of the two submissions selected. </p>",
      "votes": null,
      "replies": [
        {
          "id": 246776,
          "author_name": "addisonhoward",
          "author_url": "",
          "post_date": "11/21/2017 19:00:02",
          "content": "<p>Devin,</p>\n\n<p>Submissions for the Raspberry Pi Solution will be separately submitted and do not need to be one of your final two submissions.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 246860,
      "author_name": "barbolo",
      "author_url": "",
      "post_date": "11/21/2017 21:41:31",
      "content": "<p>I'm working on a solution compatible with the RPI3 description. My current model has 3,1MB, inference time &lt; 100ms and is achieving 76% accuracy. The model was trained for less than an hour. It's overfitting the training data, so I'll try to augment the data. Let's see what happens.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "245497": "How would special models be recognized? \n\nAssuming a model met all qualifications for the special prize, how would the person who made it know if its accuracy is \"good enough\" compared to other \"special\" models?\n\nSince it'll have a low score on the Public Leaderboard relative to other \"normal\" submissions.",
    "245524": "It would be nice, but it would require a lot of setup on Kaggle's end to automatically run code submissions on a pi, and I suspect a lot of the best pi-sized submissions will come later in the competition when people trim down their regular models.",
    "245760": "Requiring a lot of setup on Kaggle's end is the problem, though I'm interested to see what they'll come up with if they choose not to make another leaderboard or solution checker.",
    "246772": "The main question I have is whether the raspberry pi solution must be selected as one of the final two submissions, and whether it has to be our highest scoring submission of the two submissions selected.",
    "246776": "Devin,\n\nSubmissions for the Raspberry Pi Solution will be separately submitted and do not need to be one of your final two submissions.",
    "246860": "I'm working on a solution compatible with the RPI3 description. My current model has 3,1MB, inference time &lt; 100ms and is achieving 76% accuracy. The model was trained for less than an hour. It's overfitting the training data, so I'll try to augment the data. Let's see what happens."
  },
  "source": "meta"
}