{
  "id": 325430,
  "title": "Couple questions about the competition - From a beginner :)",
  "url": "/competitions/AI4Code/discussion/325430",
  "author_name": "",
  "post_date": "2022-05-16T12:46:38.146729Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>1) Is this a ML problem or DL problem? Sounds like DL to me. (In the sense that should I use a NN like a LSTM or a machine learning algorithm)<br>\n2) What is my final output exactly supposed to be? Like when my model is tested, its supposed to sort the markdown and code cells?<br>\n3) How would I go about it? If my above assumption is true, My idea is to use the md cells explanations, and generate some code out of it and then compare it to the actual code cells. The one with the highest similarity would be paired with that markdown cell.<br>\nFrom,<br>\nKrakenly, A Beginner. </p>",
  "messages": [
    {
      "id": "1791880",
      "postDate": "05/16/2022 12:46:38",
      "content": "<p>1) Is this a ML problem or DL problem? Sounds like DL to me. (In the sense that should I use a NN like a LSTM or a machine learning algorithm)<br>\n2) What is my final output exactly supposed to be? Like when my model is tested, its supposed to sort the markdown and code cells?<br>\n3) How would I go about it? If my above assumption is true, My idea is to use the md cells explanations, and generate some code out of it and then compare it to the actual code cells. The one with the highest similarity would be paired with that markdown cell.<br>\nFrom,<br>\nKrakenly, A Beginner. </p>",
      "rawMarkdown": "1) Is this a ML problem or DL problem? Sounds like DL to me. (In the sense that should I use a NN like a LSTM or a machine learning algorithm)\n2) What is my final output exactly supposed to be? Like when my model is tested, its supposed to sort the markdown and code cells?\n3) How would I go about it? If my above assumption is true, My idea is to use the md cells explanations, and generate some code out of it and then compare it to the actual code cells. The one with the highest similarity would be paired with that markdown cell.\nFrom,\nKrakenly, A Beginner.",
      "votes": null
    },
    {
      "id": "1791919",
      "postDate": "05/16/2022 13:29:46",
      "content": "<p>1) It sounds like you're on the right track with using a DL algorithm for this problem. The final output should be a sorted list of markdown and code cells.</p>\n<p>2) You could generate code from the markdown cells by extracting key phrases and matching them to existing code snippets. The one with the highest similarity would be paired with that markdown cell.</p>",
      "rawMarkdown": "1) It sounds like you're on the right track with using a DL algorithm for this problem. The final output should be a sorted list of markdown and code cells.\n\n2) You could generate code from the markdown cells by extracting key phrases and matching them to existing code snippets. The one with the highest similarity would be paired with that markdown cell.",
      "votes": null
    },
    {
      "id": "1791989",
      "postDate": "05/16/2022 14:28:00",
      "content": "<p>By a key phrase, you mean something which has a high 'value' i.e. something with a high rank (But that would require me to first rank the sentences in a markdown cell)? And then maybe train a NN that will match that key phrase into the right code? So for eg. if the phrase is <code>define df as a DataFrame with x parameters</code>, I would translate that into <code>df = Dataframe(**params)</code> I  How would I collect the data for such a NN? Should I try extracting it out of the main data?</p>",
      "rawMarkdown": "By a key phrase, you mean something which has a high 'value' i.e. something with a high rank (But that would require me to first rank the sentences in a markdown cell)? And then maybe train a NN that will match that key phrase into the right code? So for eg. if the phrase is `define df as a DataFrame with x parameters`, I would translate that into `df = Dataframe(**params)` I  How would I collect the data for such a NN? Should I try extracting it out of the main data?",
      "votes": null
    },
    {
      "id": "1791990",
      "postDate": "05/16/2022 14:29:23",
      "content": "<p>Or I could maybe compare the key phrases in md in the train data with the code snippets like you suggest? For this part I believe an LSTM would perform the best.</p>",
      "rawMarkdown": "Or I could maybe compare the key phrases in md in the train data with the code snippets like you suggest? For this part I believe an LSTM would perform the best.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1791919,
      "author_name": "",
      "author_url": "",
      "post_date": "05/16/2022 13:29:46",
      "content": "<p>1) It sounds like you're on the right track with using a DL algorithm for this problem. The final output should be a sorted list of markdown and code cells.</p>\n<p>2) You could generate code from the markdown cells by extracting key phrases and matching them to existing code snippets. The one with the highest similarity would be paired with that markdown cell.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1791989,
          "author_name": "krakenly",
          "author_url": "",
          "post_date": "05/16/2022 14:28:00",
          "content": "<p>By a key phrase, you mean something which has a high 'value' i.e. something with a high rank (But that would require me to first rank the sentences in a markdown cell)? And then maybe train a NN that will match that key phrase into the right code? So for eg. if the phrase is <code>define df as a DataFrame with x parameters</code>, I would translate that into <code>df = Dataframe(**params)</code> I  How would I collect the data for such a NN? Should I try extracting it out of the main data?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1791990,
          "author_name": "krakenly",
          "author_url": "",
          "post_date": "05/16/2022 14:29:23",
          "content": "<p>Or I could maybe compare the key phrases in md in the train data with the code snippets like you suggest? For this part I believe an LSTM would perform the best.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1791880": "1) Is this a ML problem or DL problem? Sounds like DL to me. (In the sense that should I use a NN like a LSTM or a machine learning algorithm)\n2) What is my final output exactly supposed to be? Like when my model is tested, its supposed to sort the markdown and code cells?\n3) How would I go about it? If my above assumption is true, My idea is to use the md cells explanations, and generate some code out of it and then compare it to the actual code cells. The one with the highest similarity would be paired with that markdown cell.\nFrom,\nKrakenly, A Beginner.",
    "1791919": "1) It sounds like you're on the right track with using a DL algorithm for this problem. The final output should be a sorted list of markdown and code cells.\n\n2) You could generate code from the markdown cells by extracting key phrases and matching them to existing code snippets. The one with the highest similarity would be paired with that markdown cell.",
    "1791989": "By a key phrase, you mean something which has a high 'value' i.e. something with a high rank (But that would require me to first rank the sentences in a markdown cell)? And then maybe train a NN that will match that key phrase into the right code? So for eg. if the phrase is `define df as a DataFrame with x parameters`, I would translate that into `df = Dataframe(**params)` I  How would I collect the data for such a NN? Should I try extracting it out of the main data?",
    "1791990": "Or I could maybe compare the key phrases in md in the train data with the code snippets like you suggest? For this part I believe an LSTM would perform the best."
  },
  "source": "meta"
}