{
  "id": 56001,
  "title": "Open Solution Journal [LB 0.2248]",
  "url": "/competitions/avito-demand-prediction/discussion/56001",
  "author_name": "",
  "post_date": "2018-05-04T10:05:54.992989600Z",
  "votes": 14,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hi everyone,</p>\n\n<p>I would like to start a journey of our solution to this challenge.\nAll the code, along with the current feautures/issues and project board is publicly available <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction\">github repo</a> . We will welcome any feedback and contributions so feel free to drop an Issue or Feature Request!</p>\n\n<p>Here is a bried sum up of our very first pipeline.</p>\n\n<p>Feature Extraction: </p>\n\n<ol>\n<li><p>Target Encoding of categorical variables <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L115-L151\">code</a></p></li>\n<li><p>Groupby Aggregations on some categorical features <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L236-L255\">code</a> <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/pipeline_config.py#L66-L76\">config</a></p></li>\n<li><p>Price as is</p></li>\n</ol>\n\n<p>Modeling:</p>\n\n<ol>\n<li>Ran some quick hyperparam search on LightGBM <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune_random_search.yaml#L38-L55\">config</a></li>\n</ol>\n\n<p>and got the best model with parameters <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune.yaml#L38-L55\">config</a></p>\n\n<p>Postprocessing</p>\n\n<ol>\n<li>Clip probabilities to [0-1] range </li>\n</ol>\n\n<p>You can investigate more in the following  <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/notebooks/devbook.ipynb.\">notebook</a></p>\n\n<p>Feel free to play with the code, comment and ask questions as there is a lot to explore.</p>\n\n<p>Enjoy and good luck!</p>",
  "messages": [
    {
      "id": "323076",
      "postDate": "05/04/2018 10:05:54",
      "content": "<p>Hi everyone,</p>\n\n<p>I would like to start a journey of our solution to this challenge.\nAll the code, along with the current feautures/issues and project board is publicly available <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction\">github repo</a> . We will welcome any feedback and contributions so feel free to drop an Issue or Feature Request!</p>\n\n<p>Here is a bried sum up of our very first pipeline.</p>\n\n<p>Feature Extraction: </p>\n\n<ol>\n<li><p>Target Encoding of categorical variables <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L115-L151\">code</a></p></li>\n<li><p>Groupby Aggregations on some categorical features <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L236-L255\">code</a> <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/pipeline_config.py#L66-L76\">config</a></p></li>\n<li><p>Price as is</p></li>\n</ol>\n\n<p>Modeling:</p>\n\n<ol>\n<li>Ran some quick hyperparam search on LightGBM <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune_random_search.yaml#L38-L55\">config</a></li>\n</ol>\n\n<p>and got the best model with parameters <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune.yaml#L38-L55\">config</a></p>\n\n<p>Postprocessing</p>\n\n<ol>\n<li>Clip probabilities to [0-1] range </li>\n</ol>\n\n<p>You can investigate more in the following  <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/notebooks/devbook.ipynb.\">notebook</a></p>\n\n<p>Feel free to play with the code, comment and ask questions as there is a lot to explore.</p>\n\n<p>Enjoy and good luck!</p>",
      "rawMarkdown": "Hi everyone,\n\nI would like to start a journey of our solution to this challenge.\nAll the code, along with the current feautures/issues and project board is publicly available [github repo][1] . We will welcome any feedback and contributions so feel free to drop an Issue or Feature Request!\n\nHere is a bried sum up of our very first pipeline.\n\nFeature Extraction: \n\n1. Target Encoding of categorical variables [code][2]\n\n2. Groupby Aggregations on some categorical features [code][3] [config][4]\n\n3. Price as is\n\nModeling:\n\n1. Ran some quick hyperparam search on LightGBM [config][5]\n\nand got the best model with parameters [config][6]\n\nPostprocessing\n\n1. Clip probabilities to [0-1] range \n\nYou can investigate more in the following  [notebook][7]\n\nFeel free to play with the code, comment and ask questions as there is a lot to explore.\n\nEnjoy and good luck!\n\n\n  [1]: https://github.com/minerva-ml/open-solution-avito-demand-prediction\n  [2]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L115-L151\n  [3]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L236-L255\n  [4]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/pipeline_config.py#L66-L76\n  [5]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune_random_search.yaml#L38-L55\n  [6]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune.yaml#L38-L55\n  [7]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/notebooks/devbook.ipynb.",
      "votes": null
    },
    {
      "id": "323710",
      "postDate": "05/06/2018 00:07:30",
      "content": "<p>Very nice pipeline for this competition </p>",
      "rawMarkdown": "Very nice pipeline for this competition",
      "votes": null
    },
    {
      "id": "323797",
      "postDate": "05/06/2018 08:34:32",
      "content": "<p>Thanks DUO! </p>\n\n<p>We will add text and timestamp features this week so I expect significant improvement in the score.</p>",
      "rawMarkdown": "Thanks DUO! \n\nWe will add text and timestamp features this week so I expect significant improvement in the score.",
      "votes": null
    },
    {
      "id": "325611",
      "postDate": "05/08/2018 16:18:52",
      "content": "<p>Thanks mate! I am in awe of your coding skills. You pump out code really fast :)</p>",
      "rawMarkdown": "Thanks mate! I am in awe of your coding skills. You pump out code really fast :)",
      "votes": null
    },
    {
      "id": "325745",
      "postDate": "05/08/2018 19:56:04",
      "content": "<p>Haha, thanks but we are reusing a lot from other open-solutions on <a href=\"https://github.com/minerva-ml\">https://github.com/minerva-ml</a> and steps repo <a href=\"https://github.com/minerva-ml/steps/tree/dev\">https://github.com/minerva-ml/steps/tree/dev</a> (it should be on pip by the end of the month). </p>\n\n<p>Anyhow stay tuned for updates/new features.</p>",
      "rawMarkdown": "Haha, thanks but we are reusing a lot from other open-solutions on https://github.com/minerva-ml and steps repo https://github.com/minerva-ml/steps/tree/dev (it should be on pip by the end of the month). \n\nAnyhow stay tuned for updates/new features.",
      "votes": null
    },
    {
      "id": "325889",
      "postDate": "05/09/2018 02:38:10",
      "content": "<p>cool! :)</p>",
      "rawMarkdown": "cool! :)",
      "votes": null
    },
    {
      "id": "332442",
      "postDate": "05/23/2018 07:02:53",
      "content": "<p>UPDATE</p>\n\n<p>Hi there,\nI finally had the time to make some improvements. </p>\n\n<p>So this is what changed:</p>\n\n<p>Feature Extraction:</p>\n\n<ol>\n<li><p>Added Tfidf <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L577-L602\">code</a></p></li>\n<li><p>Hand Crafted text features like len vs word counts and so on <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L248-L288\">code</a></p></li>\n<li><p>Added text over <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L291-L317\">code</a></p></li>\n<li><p>Added Image statistics features like intensity histogram and stuff. It is based on the kernel posted somewhere here. The good part is, it is parallel <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L605-L686\">code</a></p></li>\n</ol>\n\n<p>Modelling:</p>\n\n<ol>\n<li>LGBM still, just updated the hyperparameters <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune.yaml#L45-L63\">config</a></li>\n</ol>\n\n<p>I am working on:</p>\n\n<ol>\n<li><p>Adding features based on the <code>train_periods.csv</code></p></li>\n<li><p>I am training Image model that classifies <code>category_name</code> and <code>parent_category_name</code> . I will add prediction distributions over those classes as features. Also if the correct label is in top_1 and top_5 for additional signal.</p></li>\n</ol>",
      "rawMarkdown": "UPDATE\n\nHi there,\nI finally had the time to make some improvements. \n\nSo this is what changed:\n\nFeature Extraction:\n\n1. Added Tfidf [code][1]\n\n2. Hand Crafted text features like len vs word counts and so on [code][2]\n\n3. Added text over [code][3]\n\n4. Added Image statistics features like intensity histogram and stuff. It is based on the kernel posted somewhere here. The good part is, it is parallel [code][4]\n\nModelling:\n\n1. LGBM still, just updated the hyperparameters [config][5]\n\nI am working on:\n\n1. Adding features based on the `train_periods.csv`\n\n2. I am training Image model that classifies `category_name` and `parent_category_name` . I will add prediction distributions over those classes as features. Also if the correct label is in top_1 and top_5 for additional signal.\n\n  [1]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L577-L602\n  [2]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L248-L288\n  [3]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L291-L317\n  [4]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L605-L686\n[5]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune.yaml#L45-L63",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 323710,
      "author_name": "classtag",
      "author_url": "",
      "post_date": "05/06/2018 00:07:30",
      "content": "<p>Very nice pipeline for this competition </p>",
      "votes": null,
      "replies": [
        {
          "id": 323797,
          "author_name": "jakubczakon",
          "author_url": "",
          "post_date": "05/06/2018 08:34:32",
          "content": "<p>Thanks DUO! </p>\n\n<p>We will add text and timestamp features this week so I expect significant improvement in the score.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 325611,
      "author_name": "tezdhar",
      "author_url": "",
      "post_date": "05/08/2018 16:18:52",
      "content": "<p>Thanks mate! I am in awe of your coding skills. You pump out code really fast :)</p>",
      "votes": null,
      "replies": [
        {
          "id": 325745,
          "author_name": "jakubczakon",
          "author_url": "",
          "post_date": "05/08/2018 19:56:04",
          "content": "<p>Haha, thanks but we are reusing a lot from other open-solutions on <a href=\"https://github.com/minerva-ml\">https://github.com/minerva-ml</a> and steps repo <a href=\"https://github.com/minerva-ml/steps/tree/dev\">https://github.com/minerva-ml/steps/tree/dev</a> (it should be on pip by the end of the month). </p>\n\n<p>Anyhow stay tuned for updates/new features.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 325889,
          "author_name": "tezdhar",
          "author_url": "",
          "post_date": "05/09/2018 02:38:10",
          "content": "<p>cool! :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 332442,
      "author_name": "jakubczakon",
      "author_url": "",
      "post_date": "05/23/2018 07:02:53",
      "content": "<p>UPDATE</p>\n\n<p>Hi there,\nI finally had the time to make some improvements. </p>\n\n<p>So this is what changed:</p>\n\n<p>Feature Extraction:</p>\n\n<ol>\n<li><p>Added Tfidf <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L577-L602\">code</a></p></li>\n<li><p>Hand Crafted text features like len vs word counts and so on <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L248-L288\">code</a></p></li>\n<li><p>Added text over <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L291-L317\">code</a></p></li>\n<li><p>Added Image statistics features like intensity histogram and stuff. It is based on the kernel posted somewhere here. The good part is, it is parallel <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L605-L686\">code</a></p></li>\n</ol>\n\n<p>Modelling:</p>\n\n<ol>\n<li>LGBM still, just updated the hyperparameters <a href=\"https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune.yaml#L45-L63\">config</a></li>\n</ol>\n\n<p>I am working on:</p>\n\n<ol>\n<li><p>Adding features based on the <code>train_periods.csv</code></p></li>\n<li><p>I am training Image model that classifies <code>category_name</code> and <code>parent_category_name</code> . I will add prediction distributions over those classes as features. Also if the correct label is in top_1 and top_5 for additional signal.</p></li>\n</ol>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "323076": "Hi everyone,\n\nI would like to start a journey of our solution to this challenge.\nAll the code, along with the current feautures/issues and project board is publicly available [github repo][1] . We will welcome any feedback and contributions so feel free to drop an Issue or Feature Request!\n\nHere is a bried sum up of our very first pipeline.\n\nFeature Extraction: \n\n1. Target Encoding of categorical variables [code][2]\n\n2. Groupby Aggregations on some categorical features [code][3] [config][4]\n\n3. Price as is\n\nModeling:\n\n1. Ran some quick hyperparam search on LightGBM [config][5]\n\nand got the best model with parameters [config][6]\n\nPostprocessing\n\n1. Clip probabilities to [0-1] range \n\nYou can investigate more in the following  [notebook][7]\n\nFeel free to play with the code, comment and ask questions as there is a lot to explore.\n\nEnjoy and good luck!\n\n\n  [1]: https://github.com/minerva-ml/open-solution-avito-demand-prediction\n  [2]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L115-L151\n  [3]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L236-L255\n  [4]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/pipeline_config.py#L66-L76\n  [5]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune_random_search.yaml#L38-L55\n  [6]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune.yaml#L38-L55\n  [7]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/notebooks/devbook.ipynb.",
    "323710": "Very nice pipeline for this competition",
    "323797": "Thanks DUO! \n\nWe will add text and timestamp features this week so I expect significant improvement in the score.",
    "325611": "Thanks mate! I am in awe of your coding skills. You pump out code really fast :)",
    "325745": "Haha, thanks but we are reusing a lot from other open-solutions on https://github.com/minerva-ml and steps repo https://github.com/minerva-ml/steps/tree/dev (it should be on pip by the end of the month). \n\nAnyhow stay tuned for updates/new features.",
    "325889": "cool! :)",
    "332442": "UPDATE\n\nHi there,\nI finally had the time to make some improvements. \n\nSo this is what changed:\n\nFeature Extraction:\n\n1. Added Tfidf [code][1]\n\n2. Hand Crafted text features like len vs word counts and so on [code][2]\n\n3. Added text over [code][3]\n\n4. Added Image statistics features like intensity histogram and stuff. It is based on the kernel posted somewhere here. The good part is, it is parallel [code][4]\n\nModelling:\n\n1. LGBM still, just updated the hyperparameters [config][5]\n\nI am working on:\n\n1. Adding features based on the `train_periods.csv`\n\n2. I am training Image model that classifies `category_name` and `parent_category_name` . I will add prediction distributions over those classes as features. Also if the correct label is in top_1 and top_5 for additional signal.\n\n  [1]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L577-L602\n  [2]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L248-L288\n  [3]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L291-L317\n  [4]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/feature_extraction.py#L605-L686\n[5]: https://github.com/minerva-ml/open-solution-avito-demand-prediction/blob/master/neptune.yaml#L45-L63"
  },
  "source": "meta"
}