{
  "id": 229168,
  "title": "Stereochemical Layer (/t) Prediction (93% Accuracy)",
  "url": "/competitions/bms-molecular-translation/discussion/229168",
  "author_name": "",
  "post_date": "2021-03-28T17:50:20.381435900Z",
  "votes": 13,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello,</p>\n<p>I just shared my first-ever notebook on Kaggle: <a href=\"https://www.kaggle.com/sorkun/bms-mt-stereochemical-layer-prediction\" target=\"_blank\">https://www.kaggle.com/sorkun/bms-mt-stereochemical-layer-prediction</a></p>\n<p>Simply my model (binary-classifier) predicts the presence of <strong>Stereochemical Layer (/t).</strong></p>\n<p>Stereochemical Layer is represented by <strong>/t/m/s</strong> sequential tags in InChI string. If a molecule contains one of the following bonds shown in the image below, Its InChI has to contain <strong>/t</strong> tag.</p>\n<p><strong>16.5%</strong> of training data contains <strong>/t</strong> tag. Almost all also contain the following <strong>/m/s</strong> tags. (so it worths 6+ LD)</p>\n<p>I used <strong>300k</strong> of training data (balanced %50-%50) and obtained <strong>93% accuracy</strong>.</p>\n<p>Such focused models can be useful for validating InChIs generated from complete models such as captioning.</p>\n<p><img src=\"https://natefsi.weebly.com/uploads/5/4/8/7/54874551/944076.png?493\" alt=\"\"><br>\n Image from: <a href=\"https://natefsi.weebly.com/\" target=\"_blank\">https://natefsi.weebly.com/</a></p>",
  "messages": [
    {
      "id": "1255349",
      "postDate": "03/28/2021 17:50:20",
      "content": "<p>Hello,</p>\n<p>I just shared my first-ever notebook on Kaggle: <a href=\"https://www.kaggle.com/sorkun/bms-mt-stereochemical-layer-prediction\" target=\"_blank\">https://www.kaggle.com/sorkun/bms-mt-stereochemical-layer-prediction</a></p>\n<p>Simply my model (binary-classifier) predicts the presence of <strong>Stereochemical Layer (/t).</strong></p>\n<p>Stereochemical Layer is represented by <strong>/t/m/s</strong> sequential tags in InChI string. If a molecule contains one of the following bonds shown in the image below, Its InChI has to contain <strong>/t</strong> tag.</p>\n<p><strong>16.5%</strong> of training data contains <strong>/t</strong> tag. Almost all also contain the following <strong>/m/s</strong> tags. (so it worths 6+ LD)</p>\n<p>I used <strong>300k</strong> of training data (balanced %50-%50) and obtained <strong>93% accuracy</strong>.</p>\n<p>Such focused models can be useful for validating InChIs generated from complete models such as captioning.</p>\n<p><img src=\"https://natefsi.weebly.com/uploads/5/4/8/7/54874551/944076.png?493\" alt=\"\"><br>\n Image from: <a href=\"https://natefsi.weebly.com/\" target=\"_blank\">https://natefsi.weebly.com/</a></p>",
      "rawMarkdown": "Hello,\n\nI just shared my first-ever notebook on Kaggle: https://www.kaggle.com/sorkun/bms-mt-stereochemical-layer-prediction\n\nSimply my model (binary-classifier) predicts the presence of **Stereochemical Layer (/t).**\n\nStereochemical Layer is represented by **/t/m/s** sequential tags in InChI string. If a molecule contains one of the following bonds shown in the image below, Its InChI has to contain **/t** tag.\n\n**16.5%** of training data contains **/t** tag. Almost all also contain the following **/m/s** tags. (so it worths 6+ LD)\n\nI used **300k** of training data (balanced %50-%50) and obtained **93% accuracy**.\n\nSuch focused models can be useful for validating InChIs generated from complete models such as captioning.\n\n![](https://natefsi.weebly.com/uploads/5/4/8/7/54874551/944076.png?493)\n Image from: https://natefsi.weebly.com/",
      "votes": null
    },
    {
      "id": "1255922",
      "postDate": "03/29/2021 11:09:44",
      "content": "<p>An example prediction on test set:<br>\n<img src=\"https://i.imgur.com/i2uzAXs.png\" alt=\"\"></p>\n<pre><code>Predictions:\n[0.104, 0.113, 0.106, 0.105, 1.000],\n[0.106, 0.112, 0.114, 0.107, 0.105],\n[0.106, 0.112, 0.285, 0.104, 0.105],\n[0.105, 1.000, 0.105, 0.106, 0.153]\n</code></pre>",
      "rawMarkdown": "An example prediction on test set:\n![](https://i.imgur.com/i2uzAXs.png)\n\n```\nPredictions:\n[0.104, 0.113, 0.106, 0.105, 1.000],\n[0.106, 0.112, 0.114, 0.107, 0.105],\n[0.106, 0.112, 0.285, 0.104, 0.105],\n[0.105, 1.000, 0.105, 0.106, 0.153]\n```",
      "votes": null
    },
    {
      "id": "1256289",
      "postDate": "03/29/2021 18:07:40",
      "content": "<p>Your 1st Notebook is awesome. I hope other Kagglers have the opportunity to read it.</p>",
      "rawMarkdown": "Your 1st Notebook is awesome. I hope other Kagglers have the opportunity to read it.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1255922,
      "author_name": "sorkun",
      "author_url": "",
      "post_date": "03/29/2021 11:09:44",
      "content": "<p>An example prediction on test set:<br>\n<img src=\"https://i.imgur.com/i2uzAXs.png\" alt=\"\"></p>\n<pre><code>Predictions:\n[0.104, 0.113, 0.106, 0.105, 1.000],\n[0.106, 0.112, 0.114, 0.107, 0.105],\n[0.106, 0.112, 0.285, 0.104, 0.105],\n[0.105, 1.000, 0.105, 0.106, 0.153]\n</code></pre>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1256289,
      "author_name": "mpwolke",
      "author_url": "",
      "post_date": "03/29/2021 18:07:40",
      "content": "<p>Your 1st Notebook is awesome. I hope other Kagglers have the opportunity to read it.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1255349": "Hello,\n\nI just shared my first-ever notebook on Kaggle: https://www.kaggle.com/sorkun/bms-mt-stereochemical-layer-prediction\n\nSimply my model (binary-classifier) predicts the presence of **Stereochemical Layer (/t).**\n\nStereochemical Layer is represented by **/t/m/s** sequential tags in InChI string. If a molecule contains one of the following bonds shown in the image below, Its InChI has to contain **/t** tag.\n\n**16.5%** of training data contains **/t** tag. Almost all also contain the following **/m/s** tags. (so it worths 6+ LD)\n\nI used **300k** of training data (balanced %50-%50) and obtained **93% accuracy**.\n\nSuch focused models can be useful for validating InChIs generated from complete models such as captioning.\n\n![](https://natefsi.weebly.com/uploads/5/4/8/7/54874551/944076.png?493)\n Image from: https://natefsi.weebly.com/",
    "1255922": "An example prediction on test set:\n![](https://i.imgur.com/i2uzAXs.png)\n\n```\nPredictions:\n[0.104, 0.113, 0.106, 0.105, 1.000],\n[0.106, 0.112, 0.114, 0.107, 0.105],\n[0.106, 0.112, 0.285, 0.104, 0.105],\n[0.105, 1.000, 0.105, 0.106, 0.153]\n```",
    "1256289": "Your 1st Notebook is awesome. I hope other Kagglers have the opportunity to read it."
  },
  "source": "meta"
}