{
  "id": 198612,
  "title": "Creating Partial Dependence Plots ",
  "url": "/competitions/riiid-test-answer-prediction/discussion/198612",
  "author_name": "",
  "post_date": "2020-11-22T04:12:41.024849300Z",
  "votes": 3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi , I had recently learned about partial dependence plots which shows how each feature affects the prediction , however it is compatible to sckit-learn models only , but since we are using LGBM can we create the same in this case too ?? </p>",
  "messages": [
    {
      "id": "1086803",
      "postDate": "11/22/2020 04:12:41",
      "content": "<p>Hi , I had recently learned about partial dependence plots which shows how each feature affects the prediction , however it is compatible to sckit-learn models only , but since we are using LGBM can we create the same in this case too ?? </p>",
      "rawMarkdown": "Hi , I had recently learned about partial dependence plots which shows how each feature affects the prediction , however it is compatible to sckit-learn models only , but since we are using LGBM can we create the same in this case too ??",
      "votes": null
    },
    {
      "id": "1086902",
      "postDate": "11/22/2020 07:02:59",
      "content": "<p>Feature importance plots can be generated for both LGBM and Catboost - assume also XGBoost but I am not familiar with that library.</p>\n<p>LGBM can show two different importance metrics.   Off line right now or would paste some code - but most shared kernels using LGBM have importance plots.</p>",
      "rawMarkdown": "Feature importance plots can be generated for both LGBM and Catboost - assume also XGBoost but I am not familiar with that library.\n\nLGBM can show two different importance metrics.   Off line right now or would paste some code - but most shared kernels using LGBM have importance plots.",
      "votes": null
    },
    {
      "id": "1086915",
      "postDate": "11/22/2020 07:12:25",
      "content": "<p>No , feature importance tells for overall data, it doesn't iterate over every row of data , partial dependence plots show for every row of data. </p>",
      "rawMarkdown": "No , feature importance tells for overall data, it doesn't iterate over every row of data , partial dependence plots show for every row of data.",
      "votes": null
    },
    {
      "id": "1087018",
      "postDate": "11/22/2020 09:23:12",
      "content": "<p>Hmmm - you want 99,000,000 rows worth of information?  </p>",
      "rawMarkdown": "Hmmm - you want 99,000,000 rows worth of information?",
      "votes": null
    },
    {
      "id": "1087028",
      "postDate": "11/22/2020 09:28:08",
      "content": "<p>Lmao , in this case , you are right ,  it might not be possible </p>",
      "rawMarkdown": "Lmao , in this case , you are right ,  it might not be possible",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1086902,
      "author_name": "pcjimmmy",
      "author_url": "",
      "post_date": "11/22/2020 07:02:59",
      "content": "<p>Feature importance plots can be generated for both LGBM and Catboost - assume also XGBoost but I am not familiar with that library.</p>\n<p>LGBM can show two different importance metrics.   Off line right now or would paste some code - but most shared kernels using LGBM have importance plots.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1086915,
          "author_name": "sahilmaheshwari",
          "author_url": "",
          "post_date": "11/22/2020 07:12:25",
          "content": "<p>No , feature importance tells for overall data, it doesn't iterate over every row of data , partial dependence plots show for every row of data. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1087018,
          "author_name": "pcjimmmy",
          "author_url": "",
          "post_date": "11/22/2020 09:23:12",
          "content": "<p>Hmmm - you want 99,000,000 rows worth of information?  </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1087028,
          "author_name": "sahilmaheshwari",
          "author_url": "",
          "post_date": "11/22/2020 09:28:08",
          "content": "<p>Lmao , in this case , you are right ,  it might not be possible </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1086803": "Hi , I had recently learned about partial dependence plots which shows how each feature affects the prediction , however it is compatible to sckit-learn models only , but since we are using LGBM can we create the same in this case too ??",
    "1086902": "Feature importance plots can be generated for both LGBM and Catboost - assume also XGBoost but I am not familiar with that library.\n\nLGBM can show two different importance metrics.   Off line right now or would paste some code - but most shared kernels using LGBM have importance plots.",
    "1086915": "No , feature importance tells for overall data, it doesn't iterate over every row of data , partial dependence plots show for every row of data.",
    "1087018": "Hmmm - you want 99,000,000 rows worth of information?",
    "1087028": "Lmao , in this case , you are right ,  it might not be possible"
  },
  "source": "meta"
}