{
  "id": 552569,
  "title": "16th Place Solution",
  "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/552569",
  "author_name": "Jack (Japan)",
  "post_date": "2024-12-20T10:42:30.942000",
  "votes": 42,
  "comment_count": 15,
  "views": 0,
  "content": "<p>It may have been partly due to luck, but I'm happy to have made a shake-up into the gold medal range.</p>\n<p>I published my solution notebook.</p>\n<ul>\n<li><a href=\"https://www.kaggle.com/code/rsakata/cmi-piu-16th-place-solution\" target=\"_blank\">https://www.kaggle.com/code/rsakata/cmi-piu-16th-place-solution</a></li>\n</ul>\n<p>The main points of my solution are as follows:</p>\n<ul>\n<li>imputation of missing values with ItrativeImputer</li>\n<li>feature engineering from parquet files</li>\n<li>LightGBM training with custom QWK objective and metric</li>\n<li>performing 10 x 10 nested cross-validation to get reliable validation scores and stable test predictions</li>\n<li>performing threshold optimization only once using the overall predictions from the nested cross-validation.</li>\n</ul>\n<p>Please refer to my notebook for more details on the above points.<br>\nTo confirm the robustness of my solution, I changed the seed of StratifiedKFold and checked the LB scores in late submissions. The results are as follows.</p>\n<table>\n<thead>\n<tr>\n<th>Seed</th>\n<th>Public LB</th>\n<th>Private LB</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>0</td>\n<td>0.441</td>\n<td>0.468</td>\n</tr>\n<tr>\n<td>1</td>\n<td>0.442</td>\n<td>0.466</td>\n</tr>\n<tr>\n<td>2</td>\n<td>0.446</td>\n<td>0.464</td>\n</tr>\n<tr>\n<td>3</td>\n<td>0.432</td>\n<td>0.472</td>\n</tr>\n<tr>\n<td>4</td>\n<td>0.447</td>\n<td>0.465</td>\n</tr>\n<tr>\n<td>5</td>\n<td>0.435</td>\n<td>0.470</td>\n</tr>\n<tr>\n<td>6</td>\n<td>0.432</td>\n<td>0.471</td>\n</tr>\n<tr>\n<td>7</td>\n<td>0.442</td>\n<td>0.470</td>\n</tr>\n<tr>\n<td>8</td>\n<td>0.441</td>\n<td>0.470</td>\n</tr>\n<tr>\n<td>9</td>\n<td>0.446</td>\n<td>0.469</td>\n</tr>\n<tr>\n<td><strong>ave.</strong></td>\n<td><strong>0.440</strong></td>\n<td><strong>0.469</strong></td>\n</tr>\n</tbody>\n</table>\n<p>The performance seems to be relatively stable. I also conducted a brief ablation study as follows.</p>\n<table>\n<thead>\n<tr>\n<th>Description</th>\n<th>CV (nested)</th>\n<th>Public LB</th>\n<th>Private LB</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Original</td>\n<td>0.4884</td>\n<td>0.433</td>\n<td>0.470</td>\n</tr>\n<tr>\n<td>without Parquet Features</td>\n<td>0.4821</td>\n<td>0.442</td>\n<td>0.464</td>\n</tr>\n<tr>\n<td>without Missing Value Imputation</td>\n<td>0.4726</td>\n<td>0.423</td>\n<td>0.438</td>\n</tr>\n<tr>\n<td>without Parquet Features and Missing Value Imputation</td>\n<td>0.4602</td>\n<td>0.440</td>\n<td>0.412</td>\n</tr>\n<tr>\n<td>without Custom Objective and Metric</td>\n<td>0.4810</td>\n<td>0.436</td>\n<td>0.471</td>\n</tr>\n</tbody>\n</table>\n<p>It is based on the results of a single execution, but judging from these result, it seems that missing value imputation was important in this competition. Unfortunately, my custom objective and metric only contributed to the CV score. However, I plan to continue exploring the effectiveness of this idea.</p>\n<p>Thanks for reading!</p>",
  "messages": [
    {
      "id": 3076868,
      "postDate": "2024-12-20T10:42:30.943Z",
      "content": "<p>It may have been partly due to luck, but I'm happy to have made a shake-up into the gold medal range.</p>\n<p>I published my solution notebook.</p>\n<ul>\n<li><a href=\"https://www.kaggle.com/code/rsakata/cmi-piu-16th-place-solution\" target=\"_blank\">https://www.kaggle.com/code/rsakata/cmi-piu-16th-place-solution</a></li>\n</ul>\n<p>The main points of my solution are as follows:</p>\n<ul>\n<li>imputation of missing values with ItrativeImputer</li>\n<li>feature engineering from parquet files</li>\n<li>LightGBM training with custom QWK objective and metric</li>\n<li>performing 10 x 10 nested cross-validation to get reliable validation scores and stable test predictions</li>\n<li>performing threshold optimization only once using the overall predictions from the nested cross-validation.</li>\n</ul>\n<p>Please refer to my notebook for more details on the above points.<br>\nTo confirm the robustness of my solution, I changed the seed of StratifiedKFold and checked the LB scores in late submissions. The results are as follows.</p>\n<table>\n<thead>\n<tr>\n<th>Seed</th>\n<th>Public LB</th>\n<th>Private LB</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>0</td>\n<td>0.441</td>\n<td>0.468</td>\n</tr>\n<tr>\n<td>1</td>\n<td>0.442</td>\n<td>0.466</td>\n</tr>\n<tr>\n<td>2</td>\n<td>0.446</td>\n<td>0.464</td>\n</tr>\n<tr>\n<td>3</td>\n<td>0.432</td>\n<td>0.472</td>\n</tr>\n<tr>\n<td>4</td>\n<td>0.447</td>\n<td>0.465</td>\n</tr>\n<tr>\n<td>5</td>\n<td>0.435</td>\n<td>0.470</td>\n</tr>\n<tr>\n<td>6</td>\n<td>0.432</td>\n<td>0.471</td>\n</tr>\n<tr>\n<td>7</td>\n<td>0.442</td>\n<td>0.470</td>\n</tr>\n<tr>\n<td>8</td>\n<td>0.441</td>\n<td>0.470</td>\n</tr>\n<tr>\n<td>9</td>\n<td>0.446</td>\n<td>0.469</td>\n</tr>\n<tr>\n<td><strong>ave.</strong></td>\n<td><strong>0.440</strong></td>\n<td><strong>0.469</strong></td>\n</tr>\n</tbody>\n</table>\n<p>The performance seems to be relatively stable. I also conducted a brief ablation study as follows.</p>\n<table>\n<thead>\n<tr>\n<th>Description</th>\n<th>CV (nested)</th>\n<th>Public LB</th>\n<th>Private LB</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Original</td>\n<td>0.4884</td>\n<td>0.433</td>\n<td>0.470</td>\n</tr>\n<tr>\n<td>without Parquet Features</td>\n<td>0.4821</td>\n<td>0.442</td>\n<td>0.464</td>\n</tr>\n<tr>\n<td>without Missing Value Imputation</td>\n<td>0.4726</td>\n<td>0.423</td>\n<td>0.438</td>\n</tr>\n<tr>\n<td>without Parquet Features and Missing Value Imputation</td>\n<td>0.4602</td>\n<td>0.440</td>\n<td>0.412</td>\n</tr>\n<tr>\n<td>without Custom Objective and Metric</td>\n<td>0.4810</td>\n<td>0.436</td>\n<td>0.471</td>\n</tr>\n</tbody>\n</table>\n<p>It is based on the results of a single execution, but judging from these result, it seems that missing value imputation was important in this competition. Unfortunately, my custom objective and metric only contributed to the CV score. However, I plan to continue exploring the effectiveness of this idea.</p>\n<p>Thanks for reading!</p>",
      "rawMarkdown": "It may have been partly due to luck, but I'm happy to have made a shake-up into the gold medal range.\n\nI published my solution notebook.\n- https://www.kaggle.com/code/rsakata/cmi-piu-16th-place-solution\n\nThe main points of my solution are as follows:\n- imputation of missing values with ItrativeImputer\n- feature engineering from parquet files\n- LightGBM training with custom QWK objective and metric\n- performing 10 x 10 nested cross-validation to get reliable validation scores and stable test predictions\n- performing threshold optimization only once using the overall predictions from the nested cross-validation.\n\nPlease refer to my notebook for more details on the above points.\nTo confirm the robustness of my solution, I changed the seed of StratifiedKFold and checked the LB scores in late submissions. The results are as follows.\n\n| Seed  | Public LB | Private LB |\n| --- | --- | --- |\n|0| 0.441 | 0.468 |\n|1| 0.442 | 0.466 |\n|2| 0.446 | 0.464 |\n|3| 0.432 | 0.472 |\n|4| 0.447 | 0.465 |\n|5| 0.435 | 0.470 |\n|6| 0.432 | 0.471 |\n|7| 0.442 | 0.470 |\n|8| 0.441 | 0.470 |\n|9| 0.446 | 0.469 |\n| **ave.** | **0.440** | **0.469** |  \n\nThe performance seems to be relatively stable. I also conducted a brief ablation study as follows.\n\n| Description  | CV (nested) | Public LB | Private LB |\n| --- | --- | --- | --- |\n| Original | 0.4884 | 0.433 | 0.470 |\n| without Parquet Features | 0.4821 | 0.442 | 0.464 |\n| without Missing Value Imputation | 0.4726 | 0.423 | 0.438 |\n| without Parquet Features and Missing Value Imputation | 0.4602 | 0.440 | 0.412 |\n| without Custom Objective and Metric | 0.4810 | 0.436 | 0.471 |\n\nIt is based on the results of a single execution, but judging from these result, it seems that missing value imputation was important in this competition. Unfortunately, my custom objective and metric only contributed to the CV score. However, I plan to continue exploring the effectiveness of this idea.\n\nThanks for reading!",
      "votes": 41
    },
    {
      "id": 3077475,
      "postDate": "2024-12-21T02:45:48.513Z",
      "content": "<p><a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a> <br>\nCongratulations and thanks for sharing.<br>\nI have two questions.</p>\n<ol>\n<li><p>imputation <br>\nWhat kind of reasoning led to filling in the missing values? Some may argue that the fact that the data is missing itself is valuable information and should not be filled in. Especially since LightGBM can train without handling missing values.</p></li>\n<li><p>number of folds.<br>\nI think large numbers of folds may lead overfit to validation data especially in small data, but does the nested CV prevent this ? Why do you choose 10folds?</p></li>\n</ol>",
      "rawMarkdown": "@rsakata \nCongratulations and thanks for sharing.\nI have two questions.\n\n1. imputation \nWhat kind of reasoning led to filling in the missing values? Some may argue that the fact that the data is missing itself is valuable information and should not be filled in. Especially since LightGBM can train without handling missing values.\n\n2. number of folds.\nI think large numbers of folds may lead overfit to validation data especially in small data, but does the nested CV prevent this ? Why do you choose 10folds?\n\n\n\n",
      "votes": 1,
      "replies": [
        {
          "id": 3078850,
          "postDate": "2024-12-22T23:18:07.233Z",
          "content": "<p>Thank you for you comment!</p>\n<ol>\n<li>Indeed, as you mentioned, I don't have a clear perspective on the reason of the effectiveness of missing value imputation either. However, I think that when the missing feature is strongly correlated with the target (in this competition, for instance, PreInt_EduHx-computerinternet_hoursday), it might have been better to impute the missing values rather than indirectly predicting the target from other features.</li>\n<li>Yes. Since no optimization was performed on the test data for each fold, I believe there is no risk of overfitting by increasing the number of folds.</li>\n</ol>",
          "rawMarkdown": "Thank you for you comment!\n1. Indeed, as you mentioned, I don't have a clear perspective on the reason of the effectiveness of missing value imputation either. However, I think that when the missing feature is strongly correlated with the target (in this competition, for instance, PreInt_EduHx-computerinternet_hoursday), it might have been better to impute the missing values rather than indirectly predicting the target from other features.\n2. Yes. Since no optimization was performed on the test data for each fold, I believe there is no risk of overfitting by increasing the number of folds.",
          "votes": 2
        }
      ]
    },
    {
      "id": 3077237,
      "postDate": "2024-12-20T17:42:08.363Z",
      "content": "<p>congrats <a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a> !</p>",
      "rawMarkdown": "congrats @rsakata !",
      "votes": 1
    },
    {
      "id": 3076926,
      "postDate": "2024-12-20T11:54:03.907Z",
      "content": "<p>Congratulations on the gold medal.<br>\nThanks for sharing. Great solution!</p>",
      "rawMarkdown": "Congratulations on the gold medal.\nThanks for sharing. Great solution!",
      "votes": 1
    },
    {
      "id": 3076924,
      "postDate": "2024-12-20T11:53:23.460Z",
      "content": "<p>Congratulations for the result <a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a> <br>\nKeep up the great work!</p>",
      "rawMarkdown": "Congratulations for the result @rsakata \nKeep up the great work!",
      "votes": 1
    },
    {
      "id": 3076886,
      "postDate": "2024-12-20T11:15:34.083Z",
      "content": "<p>Thank you so much for sharing <a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a>. I really liked your other notebook as well <a href=\"https://www.kaggle.com/code/rsakata/cmi-piu-optimize-qwk-by-lgb\" target=\"_blank\">https://www.kaggle.com/code/rsakata/cmi-piu-optimize-qwk-by-lgb</a>. </p>",
      "rawMarkdown": "Thank you so much for sharing @rsakata. I really liked your other notebook as well https://www.kaggle.com/code/rsakata/cmi-piu-optimize-qwk-by-lgb. ",
      "votes": 1
    },
    {
      "id": 3076936,
      "postDate": "2024-12-20T12:08:20.480Z",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a>, congrats on the gold!</p>\n<p>I'm curious, what made you decide to make only two subs? Did you expect a big shakeup from the beginning of the competition?</p>",
      "rawMarkdown": "Hi @rsakata, congrats on the gold!\n\nI'm curious, what made you decide to make only two subs? Did you expect a big shakeup from the beginning of the competition?",
      "votes": 2,
      "replies": [
        {
          "id": 3077194,
          "postDate": "2024-12-20T16:52:18.417Z",
          "content": "<p>This guy is simply unique. His name is legendary. Reaching a very high Private Rank from a couple of early submissions is his proprietary method. If I say he did it about 12 times already in various competitions, my guess will be not far from real. True GM as is.</p>",
          "rawMarkdown": "This guy is simply unique. His name is legendary. Reaching a very high Private Rank from a couple of early submissions is his proprietary method. If I say he did it about 12 times already in various competitions, my guess will be not far from real. True GM as is.",
          "votes": 2,
          "replies": [
            {
              "id": 3078851,
              "postDate": "2024-12-22T23:21:11.367Z",
              "rawMarkdown": "",
              "isDeleted": true
            }
          ]
        },
        {
          "id": 3078852,
          "postDate": "2024-12-22T23:21:32.187Z",
          "content": "<p>Thank you for your comment!  I am honored.<br>\nI determined that the information on the Public Leaderboard was completely unreliable, so I focused solely on improving the model using local validation. As a result, I ended up submitting only two submissions.</p>",
          "rawMarkdown": "Thank you for your comment!  I am honored.\nI determined that the information on the Public Leaderboard was completely unreliable, so I focused solely on improving the model using local validation. As a result, I ended up submitting only two submissions.",
          "votes": 1,
          "replies": [
            {
              "id": 3078922,
              "postDate": "2024-12-23T02:42:11.763Z",
              "content": "<p>Thank you, Jack!</p>",
              "rawMarkdown": "Thank you, Jack!",
              "votes": 1
            }
          ]
        }
      ]
    },
    {
      "id": 3084900,
      "postDate": "2024-12-31T12:56:08.707Z",
      "content": "<p><a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a> damn, very interesting!</p>",
      "rawMarkdown": "@rsakata damn, very interesting!"
    },
    {
      "id": 3079788,
      "postDate": "2024-12-24T06:10:51.917Z",
      "content": "<p>Congratulations! Looking forward to learn from you</p>",
      "rawMarkdown": "Congratulations! Looking forward to learn from you"
    },
    {
      "id": 3077106,
      "postDate": "2024-12-20T15:11:21.307Z",
      "content": "<p>Hello All,</p>\n<p>I am sorry to bother you. Since the competition ended, I will be glad if someone can assist. I started working on Kaggle competition last month. My first competition was Child Mind Institute - Problematic Internet Use and I was not able to submit my notebook. I was getting the error \"Notebook three exception\" and the submissions would fail with no score. I turned off internet but I was still getting the same error. Am I missing another thing apart from turning internet off? Or something is wrong with my code at the end part?</p>\n<p>I started another competition and I am having the same issue. Please, I will appreciate if you can help me identify what I am doing wrong. Find attached one of my notebooks for Child Mind. You can submit my notebook if you want to.</p>\n<p>I appreciate your assistance, thanks.</p>",
      "rawMarkdown": "Hello All,\n\nI am sorry to bother you. Since the competition ended, I will be glad if someone can assist. I started working on Kaggle competition last month. My first competition was Child Mind Institute - Problematic Internet Use and I was not able to submit my notebook. I was getting the error \"Notebook three exception\" and the submissions would fail with no score. I turned off internet but I was still getting the same error. Am I missing another thing apart from turning internet off? Or something is wrong with my code at the end part?\n\nI started another competition and I am having the same issue. Please, I will appreciate if you can help me identify what I am doing wrong. Find attached one of my notebooks for Child Mind. You can submit my notebook if you want to.\n\nI appreciate your assistance, thanks."
    },
    {
      "id": 3076876,
      "postDate": "2024-12-20T10:54:46.967Z",
      "content": "<p>Brilliant. Thanks for sharing!</p>",
      "rawMarkdown": "Brilliant. Thanks for sharing!",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 3077475,
      "author_name": "Aurora_blue",
      "author_url": "",
      "post_date": "2024-12-21T02:45:48.513000",
      "content": "<p><a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a> <br>\nCongratulations and thanks for sharing.<br>\nI have two questions.</p>\n<ol>\n<li><p>imputation <br>\nWhat kind of reasoning led to filling in the missing values? Some may argue that the fact that the data is missing itself is valuable information and should not be filled in. Especially since LightGBM can train without handling missing values.</p></li>\n<li><p>number of folds.<br>\nI think large numbers of folds may lead overfit to validation data especially in small data, but does the nested CV prevent this ? Why do you choose 10folds?</p></li>\n</ol>",
      "votes": 1,
      "replies": [
        {
          "id": 3078850,
          "author_name": "Jack (Japan)",
          "author_url": "",
          "post_date": "2024-12-22T23:18:07.233000",
          "content": "<p>Thank you for you comment!</p>\n<ol>\n<li>Indeed, as you mentioned, I don't have a clear perspective on the reason of the effectiveness of missing value imputation either. However, I think that when the missing feature is strongly correlated with the target (in this competition, for instance, PreInt_EduHx-computerinternet_hoursday), it might have been better to impute the missing values rather than indirectly predicting the target from other features.</li>\n<li>Yes. Since no optimization was performed on the test data for each fold, I believe there is no risk of overfitting by increasing the number of folds.</li>\n</ol>",
          "votes": 2,
          "replies": []
        }
      ]
    },
    {
      "id": 3077237,
      "author_name": "Octavi Grau",
      "author_url": "",
      "post_date": "2024-12-20T17:42:08.363000",
      "content": "<p>congrats <a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a> !</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 3076926,
      "author_name": "Solo Diver",
      "author_url": "",
      "post_date": "2024-12-20T11:54:03.907000",
      "content": "<p>Congratulations on the gold medal.<br>\nThanks for sharing. Great solution!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 3076924,
      "author_name": "Ravi Ramakrishnan",
      "author_url": "",
      "post_date": "2024-12-20T11:53:23.460000",
      "content": "<p>Congratulations for the result <a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a> <br>\nKeep up the great work!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 3076886,
      "author_name": "JamshaidSohail",
      "author_url": "",
      "post_date": "2024-12-20T11:15:34.083000",
      "content": "<p>Thank you so much for sharing <a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a>. I really liked your other notebook as well <a href=\"https://www.kaggle.com/code/rsakata/cmi-piu-optimize-qwk-by-lgb\" target=\"_blank\">https://www.kaggle.com/code/rsakata/cmi-piu-optimize-qwk-by-lgb</a>. </p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 3076936,
      "author_name": "ducnh279",
      "author_url": "",
      "post_date": "2024-12-20T12:08:20.480000",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a>, congrats on the gold!</p>\n<p>I'm curious, what made you decide to make only two subs? Did you expect a big shakeup from the beginning of the competition?</p>",
      "votes": 2,
      "replies": [
        {
          "id": 3077194,
          "author_name": "Vladimir Demidov",
          "author_url": "",
          "post_date": "2024-12-20T16:52:18.417000",
          "content": "<p>This guy is simply unique. His name is legendary. Reaching a very high Private Rank from a couple of early submissions is his proprietary method. If I say he did it about 12 times already in various competitions, my guess will be not far from real. True GM as is.</p>",
          "votes": 2,
          "replies": [
            {
              "id": 3078851,
              "author_name": "",
              "author_url": "",
              "post_date": "2024-12-22T23:21:11.367000",
              "content": "",
              "votes": 0,
              "replies": []
            }
          ]
        },
        {
          "id": 3078852,
          "author_name": "Jack (Japan)",
          "author_url": "",
          "post_date": "2024-12-22T23:21:32.187000",
          "content": "<p>Thank you for your comment!  I am honored.<br>\nI determined that the information on the Public Leaderboard was completely unreliable, so I focused solely on improving the model using local validation. As a result, I ended up submitting only two submissions.</p>",
          "votes": 1,
          "replies": [
            {
              "id": 3078922,
              "author_name": "ducnh279",
              "author_url": "",
              "post_date": "2024-12-23T02:42:11.763000",
              "content": "<p>Thank you, Jack!</p>",
              "votes": 1,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 3084900,
      "author_name": "Volodymyr Pivoshenko 🇺🇦",
      "author_url": "",
      "post_date": "2024-12-31T12:56:08.707000",
      "content": "<p><a href=\"https://www.kaggle.com/rsakata\" target=\"_blank\">@rsakata</a> damn, very interesting!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 3079788,
      "author_name": "Tanishk Patil",
      "author_url": "",
      "post_date": "2024-12-24T06:10:51.917000",
      "content": "<p>Congratulations! Looking forward to learn from you</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 3077106,
      "author_name": "kaylascho Lawrene",
      "author_url": "",
      "post_date": "2024-12-20T15:11:21.307000",
      "content": "<p>Hello All,</p>\n<p>I am sorry to bother you. Since the competition ended, I will be glad if someone can assist. I started working on Kaggle competition last month. My first competition was Child Mind Institute - Problematic Internet Use and I was not able to submit my notebook. I was getting the error \"Notebook three exception\" and the submissions would fail with no score. I turned off internet but I was still getting the same error. Am I missing another thing apart from turning internet off? Or something is wrong with my code at the end part?</p>\n<p>I started another competition and I am having the same issue. Please, I will appreciate if you can help me identify what I am doing wrong. Find attached one of my notebooks for Child Mind. You can submit my notebook if you want to.</p>\n<p>I appreciate your assistance, thanks.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 3076876,
      "author_name": "bsmelbs",
      "author_url": "",
      "post_date": "2024-12-20T10:54:46.967000",
      "content": "<p>Brilliant. Thanks for sharing!</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3076868": "It may have been partly due to luck, but I'm happy to have made a shake-up into the gold medal range.\n\nI published my solution notebook.\n- https://www.kaggle.com/code/rsakata/cmi-piu-16th-place-solution\n\nThe main points of my solution are as follows:\n- imputation of missing values with ItrativeImputer\n- feature engineering from parquet files\n- LightGBM training with custom QWK objective and metric\n- performing 10 x 10 nested cross-validation to get reliable validation scores and stable test predictions\n- performing threshold optimization only once using the overall predictions from the nested cross-validation.\n\nPlease refer to my notebook for more details on the above points.\nTo confirm the robustness of my solution, I changed the seed of StratifiedKFold and checked the LB scores in late submissions. The results are as follows.\n\n| Seed  | Public LB | Private LB |\n| --- | --- | --- |\n|0| 0.441 | 0.468 |\n|1| 0.442 | 0.466 |\n|2| 0.446 | 0.464 |\n|3| 0.432 | 0.472 |\n|4| 0.447 | 0.465 |\n|5| 0.435 | 0.470 |\n|6| 0.432 | 0.471 |\n|7| 0.442 | 0.470 |\n|8| 0.441 | 0.470 |\n|9| 0.446 | 0.469 |\n| **ave.** | **0.440** | **0.469** |  \n\nThe performance seems to be relatively stable. I also conducted a brief ablation study as follows.\n\n| Description  | CV (nested) | Public LB | Private LB |\n| --- | --- | --- | --- |\n| Original | 0.4884 | 0.433 | 0.470 |\n| without Parquet Features | 0.4821 | 0.442 | 0.464 |\n| without Missing Value Imputation | 0.4726 | 0.423 | 0.438 |\n| without Parquet Features and Missing Value Imputation | 0.4602 | 0.440 | 0.412 |\n| without Custom Objective and Metric | 0.4810 | 0.436 | 0.471 |\n\nIt is based on the results of a single execution, but judging from these result, it seems that missing value imputation was important in this competition. Unfortunately, my custom objective and metric only contributed to the CV score. However, I plan to continue exploring the effectiveness of this idea.\n\nThanks for reading!",
    "3077475": "@rsakata \nCongratulations and thanks for sharing.\nI have two questions.\n\n1. imputation \nWhat kind of reasoning led to filling in the missing values? Some may argue that the fact that the data is missing itself is valuable information and should not be filled in. Especially since LightGBM can train without handling missing values.\n\n2. number of folds.\nI think large numbers of folds may lead overfit to validation data especially in small data, but does the nested CV prevent this ? Why do you choose 10folds?\n\n\n\n",
    "3077237": "congrats @rsakata !",
    "3076926": "Congratulations on the gold medal.\nThanks for sharing. Great solution!",
    "3076924": "Congratulations for the result @rsakata \nKeep up the great work!",
    "3076886": "Thank you so much for sharing @rsakata. I really liked your other notebook as well https://www.kaggle.com/code/rsakata/cmi-piu-optimize-qwk-by-lgb. ",
    "3076936": "Hi @rsakata, congrats on the gold!\n\nI'm curious, what made you decide to make only two subs? Did you expect a big shakeup from the beginning of the competition?",
    "3084900": "@rsakata damn, very interesting!",
    "3079788": "Congratulations! Looking forward to learn from you",
    "3077106": "Hello All,\n\nI am sorry to bother you. Since the competition ended, I will be glad if someone can assist. I started working on Kaggle competition last month. My first competition was Child Mind Institute - Problematic Internet Use and I was not able to submit my notebook. I was getting the error \"Notebook three exception\" and the submissions would fail with no score. I turned off internet but I was still getting the same error. Am I missing another thing apart from turning internet off? Or something is wrong with my code at the end part?\n\nI started another competition and I am having the same issue. Please, I will appreciate if you can help me identify what I am doing wrong. Find attached one of my notebooks for Child Mind. You can submit my notebook if you want to.\n\nI appreciate your assistance, thanks.",
    "3076876": "Brilliant. Thanks for sharing!"
  }
}